cassdlcm/mediapipe
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
1,011 Commits 10 Branches 56 Tags 829 MiB
c79350da5f
Commit Graph

960 Commits

Author SHA1 Message Date
Camillo Lugaresi
bacbac8d92 Use kUtilityFramebuffer in ReadTexture 
This avoids creating a temporary framebuffer each time.

PiperOrigin-RevId: 490163892
 2022-11-21 23:59:44 -08:00
Camillo Lugaresi
a8b7761022 Define a kUtilityFramebuffer context attachment 
A framebuffer object is often needed to render to a texture or read data from it. Currently we create one in each GlCalculatorHelper, but that is redundant (we only need one per context, and multiple calculators can share the same context). Other times, the code that needs to use this doesn't own a helper. For both reasons, this should be attached to the context.

We could just make this a member of GlContext since it's so common. However, I figured we might as well use the attachment system.

PiperOrigin-RevId: 490160214
 2022-11-21 23:29:27 -08:00
Camillo Lugaresi
54a684717f Internal change 
PiperOrigin-RevId: 490159674
 2022-11-21 23:24:16 -08:00
MediaPipe Team
d43d0ff615 Internal change 
PiperOrigin-RevId: 490089940
 2022-11-21 15:47:45 -08:00
MediaPipe Team
adddf2c2ab Extracted common test helper functions out from the unittest into a sharable 
library. Also migrated away from OpenCVX.

PiperOrigin-RevId: 490074410
 2022-11-21 14:39:56 -08:00
MediaPipe Team
652423a23d Internal change 
PiperOrigin-RevId: 490053179
 2022-11-21 13:06:47 -08:00
MediaPipe Team
7f0134eecb Internal change 
PiperOrigin-RevId: 490041386
 2022-11-21 12:16:00 -08:00
MediaPipe Team
7acbf557a1 Cleanup after migration to new classification output format. 
PiperOrigin-RevId: 489921603
 2022-11-21 01:57:05 -08:00
MediaPipe Team
13c6b9a8c6 Allow kernel cache path to be specified without trailing path delimiter 
PiperOrigin-RevId: 489891079
 2022-11-20 22:20:18 -08:00
Camillo Lugaresi
3ac7f6a216 Simplify image creation in PacketCreator 
Use more existing functions, remove redundant code, remove direct use of RuntimeException.

PiperOrigin-RevId: 489868983
 2022-11-20 19:32:21 -08:00
MediaPipe Team
6cf464636b Internal change 
PiperOrigin-RevId: 489842199
 2022-11-20 14:26:22 -08:00
Scott Zhu
a367753eda Internal change 
PiperOrigin-RevId: 489824381
 2022-11-20 10:41:20 -08:00
MediaPipe Team
bdf4078e89 Internal change 
PiperOrigin-RevId: 489752009
 2022-11-19 21:14:24 -08:00
Camillo Lugaresi
a33cb1e05e Check that Java buffer supports direct access before using it 
If the buffer is not created with allocateDirect, JNI APIs will return a data pointer of nullptr and a capacity of -1. This can cause a crash when we access it.

Also clean up the code to raise exceptions instead of just logging errors and returning nullptr.

PiperOrigin-RevId: 489751312
 2022-11-19 21:04:48 -08:00
MediaPipe Team
977ee4272e Add public visibility to the model maker public API. 
PiperOrigin-RevId: 489701768
 2022-11-19 10:53:21 -08:00
Sebastian Schmidt
bbcbd5fc6c Audio Embedder for Web 
PiperOrigin-RevId: 489669966
 2022-11-19 04:49:24 -08:00
Sebastian Schmidt
e853f04b79 Create AudioTaskRunner 
PiperOrigin-RevId: 489613573
 2022-11-18 19:55:08 -08:00
Camillo Lugaresi
eb8ef1ace0 Use shared_from_this in GlTextureBuffer::GetReadView, GetWriteView 
This ensures that the callbacks in GlTextureView won't call an expired object, even if user code holds a GlTextureView after releasing the buffer.

Note that GlTextureBuffer is not always held by a shared_ptr, but it always is when GpuBuffer calls GetRead/WriteView on it. An alternative solution would have been to have GpuBuffer pass its shared_ptr<GlTextureBuffer> to the view method, which could have been implemented with some compile-time logic to detect whether the method expects such an argument. However, that doesn't seem necessary.

PiperOrigin-RevId: 489611843
 2022-11-18 19:42:34 -08:00
MediaPipe Team
bbd5da7971 Added the gray scale image support for the ImageToTensorCalculator on CPU. 
PiperOrigin-RevId: 489593917
 2022-11-18 17:12:57 -08:00
MediaPipe Team
524ac3ca61 Internal change for Model Maker 
PiperOrigin-RevId: 489540387
 2022-11-18 12:48:30 -08:00
Jiuqiang Tang
1b594a0310 Return error status when any tflite input and output tensor doesn't have valid dimensionality information that is needed to allocate Gl/Metal buffer before calling ModifyGraphWithDelegate. 
PiperOrigin-RevId: 489539740
 2022-11-18 12:44:55 -08:00
Sebastian Schmidt
71ae496a20 Add AudioEmbedder documentation 
PiperOrigin-RevId: 489532283
 2022-11-18 12:12:41 -08:00
MediaPipe Team
e2052a6a51 Rename embedding postprocessor "configure" method for consistency with classification postprocessor. 
PiperOrigin-RevId: 489518257
 2022-11-18 11:13:10 -08:00
Adam Cozzette
ac212c1507 Internal change 
PiperOrigin-RevId: 489502255
 2022-11-18 10:08:26 -08:00
MediaPipe Team
03d388fecf Add hand landmark named index constants 
PiperOrigin-RevId: 489498248
 2022-11-18 09:52:44 -08:00
MediaPipe Team
2f361e2f47 Internal change 
PiperOrigin-RevId: 489486417
 2022-11-18 08:53:43 -08:00
Jiuqiang Tang
e046982a3c Internal change 
PiperOrigin-RevId: 489484898
 2022-11-18 08:47:09 -08:00
Hadon Nash
ae44012c0c Allowing BypassCalculator to accept InputSidePackets. 
PiperOrigin-RevId: 489483992
 2022-11-18 08:43:02 -08:00
Camillo Lugaresi
efcdedbd59 Remove redundant _ios targets 
PiperOrigin-RevId: 489355333
 2022-11-17 18:16:58 -08:00
Copybara-Service
cd056e114a Merge pull request #3866 from kinaryml:fix-tasks-api-python-311-errors 
PiperOrigin-RevId: 489346167
 2022-11-17 17:20:23 -08:00
MediaPipe Team
6f3cb340e1 Internal change 
PiperOrigin-RevId: 489345940
 2022-11-17 17:16:25 -08:00
MediaPipe Team
ab3a5f0fbf Make MuxCalculator with DefaultInputStreamHandler to handle graph closure gracefully 
PiperOrigin-RevId: 489336722
 2022-11-17 16:30:35 -08:00
MediaPipe Team
a7bd725e65 Internal change 
PiperOrigin-RevId: 489331826
 2022-11-17 16:09:15 -08:00
MediaPipe Team
1fb0902aa0 Update gesture_recognizer test 
PiperOrigin-RevId: 489301508
 2022-11-17 14:04:30 -08:00
Sebastian Schmidt
3ccf7308e0 Add shared options for Text and Audio Tasks 
PiperOrigin-RevId: 489186644
 2022-11-17 05:28:27 -08:00
Hadon Nash
ea4989b6f1 Internal change 
PiperOrigin-RevId: 489135553
 2022-11-17 00:07:59 -08:00
kinaryml
5a6837d034 Fix errors that will occur in python 3.11 2022-11-16 22:08:52 -08:00
Sebastian Schmidt
899c87466e Add MP Tasks entrypoints 
PiperOrigin-RevId: 489110875
 2022-11-16 20:57:33 -08:00
Camillo Lugaresi
6fc277ee1c Internal change 
PiperOrigin-RevId: 489091534
 2022-11-16 18:38:03 -08:00
Sebastian Schmidt
e66e88802c Change NPM Bundle to ESM 
PiperOrigin-RevId: 489091370
 2022-11-16 18:34:53 -08:00
MediaPipe Team
90eb4a19d8 Internal change 
PiperOrigin-RevId: 489088227
 2022-11-16 18:16:32 -08:00
MediaPipe Team
b6b72d5e4e Add MuxCalculator test case where graph is being closed while SELECT has not been received. 
PiperOrigin-RevId: 489061902
 2022-11-16 15:56:43 -08:00
Sebastian Schmidt
3cdf0f6536 Fix a crash that occurred when a model returns fewer vector elements than before 
PiperOrigin-RevId: 489041814
 2022-11-16 14:38:37 -08:00
MediaPipe Team
74474d859e Update image_classifier demo with new ImageClassifierOption changes 
PiperOrigin-RevId: 489031381
 2022-11-16 14:00:39 -08:00
MediaPipe Team
cdd44e77b7 Internal change 
PiperOrigin-RevId: 488969539
 2022-11-16 10:07:40 -08:00
MediaPipe Team
dc9578d226 Internal change 
PiperOrigin-RevId: 488946809
 2022-11-16 08:30:20 -08:00
Sebastian Schmidt
f7aef677fc Add running mode to all vision tasks 
PiperOrigin-RevId: 488816785
 2022-11-15 19:02:21 -08:00
Camillo Lugaresi
febfc2029b Annotate plane argument 
PiperOrigin-RevId: 488813363
 2022-11-15 18:53:09 -08:00
Camillo Lugaresi
63e2089639 Deprecate a bunch of old stuff in GlCalculatorHelper 
PiperOrigin-RevId: 488813296
 2022-11-15 18:51:56 -08:00
Camillo Lugaresi
1979801a92 Remove GlCalculatorHelperImpl; merge with GlCalculatorHelper 
Originally, there were multiple implementations of GlCalculatorHelperImpl, depending on the platform and underlying GL APIs. These have all been refactored into other components, and the remaining code in this class is unified and much reduced in size. We can get rid of this implementation detail now.

PiperOrigin-RevId: 488813220
 2022-11-15 18:50:45 -08:00
Camillo Lugaresi
8b319e963a Add comment explaining ViewProvider 
This was only documented via examples (e.g. ViewProvider<GlTextureView>), but it's better to explain it properly in the header where the base case is defined.

PiperOrigin-RevId: 488813144
 2022-11-15 18:48:54 -08:00
Camillo Lugaresi
a28ccb0964 Remove unnecessary forward declarations 
PiperOrigin-RevId: 488813066
 2022-11-15 18:43:36 -08:00
Camillo Lugaresi
13b4b825d7 Remove std::shared_ptr<GpuBuffer> argument from GetRead/WriteView 
PiperOrigin-RevId: 488813004
 2022-11-15 18:41:53 -08:00
Camillo Lugaresi
1c0a1d0aab Remove shared_ptr<GpuBuffer> member from GlTextureView 
This only exists to support GlTexture's GetFrame API. It can be moved into GlTexture.

PiperOrigin-RevId: 488812896
 2022-11-15 18:40:15 -08:00
Camillo Lugaresi
767cc2ee3c More comments on gpu_buffer_storage 
This gives a basic explanation of the role of storages and views, and provides some details on how to implement a new storage type.

PiperOrigin-RevId: 488812807
 2022-11-15 18:38:29 -08:00
Camillo Lugaresi
4c874fe4cd Allow conversion of GlTextureBuffer to CVPixelBufferRef 
This means that, if an iOS application sends in a GlTextureBuffer but expects a CVPixelBufferRef as output, everything will work even if the graph just forwards the same input. Also, access by Metal calculators will also work transparently.

PiperOrigin-RevId: 488812748
 2022-11-15 18:36:45 -08:00
Camillo Lugaresi
fe66de3714 Internal change 
PiperOrigin-RevId: 488812677
 2022-11-15 18:32:54 -08:00
MediaPipe Team
2f77bf44e3 Use train_data to evaluate accuracy of unit test for gesture_recognizer due to limited dataset size. 
PiperOrigin-RevId: 488808942
 2022-11-15 18:10:28 -08:00
Camillo Lugaresi
b308c0dd5e Implement CVPixelBufferRef access as a view. 
PiperOrigin-RevId: 488798216
 2022-11-15 17:18:07 -08:00
Camillo Lugaresi
4bda012bba Factor out gl_texture_util 
PiperOrigin-RevId: 488797985
 2022-11-15 17:13:22 -08:00
Camillo Lugaresi
77b3edbb67 Internal change 
PiperOrigin-RevId: 488797407
 2022-11-15 17:09:43 -08:00
Camillo Lugaresi
7e19bbe35c Internal change 
PiperOrigin-RevId: 488795920
 2022-11-15 17:01:33 -08:00
Camillo Lugaresi
1beca61650 Register GlTextureBuffer pool with GpuBuffer 
First crack at hooking up pools with the GpuBufferStorage system. Will most likely be superseded later, but for now this works with minimal code impact: just overwrite the factory for a storage type with one that uses the pool.

PiperOrigin-RevId: 488783854
 2022-11-15 16:19:48 -08:00
Camillo Lugaresi
583d27636b Factor out ReusablePool 
PiperOrigin-RevId: 488783477
 2022-11-15 16:18:20 -08:00
Camillo Lugaresi
53d015af08 Generic MultiPool template 
PiperOrigin-RevId: 488783176
 2022-11-15 16:15:42 -08:00
Camillo Lugaresi
b9fa2e3496 Make it possible to override the SimplePool factory used by MultiPool 
This means MultiPool no longer needs a SetFlushPlatformCaches method, which was too specific to the CVPixelBufferPool.

PiperOrigin-RevId: 488783003
 2022-11-15 16:14:27 -08:00
Camillo Lugaresi
267476657d MultiPool options header refactoring 
Passing MultiPool options to the base pool factories means that we don't have to specialize which options we pass to them.

PiperOrigin-RevId: 488782861
 2022-11-15 16:13:05 -08:00
Camillo Lugaresi
7ef3185ecb Allow customizing MultiPool options 
These don't need to be constants.

PiperOrigin-RevId: 488782713
 2022-11-15 16:11:54 -08:00
Camillo Lugaresi
f13903b7c5 Call SimplePool methods directly 
This removes redundant helper functions in GpuBufferMultiPool.

PiperOrigin-RevId: 488782516
 2022-11-15 16:10:34 -08:00
Camillo Lugaresi
0c4522cb9f Move flush hook to CvPixelBufferPoolWrapper constructor 
This unifies the implementation of GpuBufferMultiPool::GetBufferFromSimplePool.

PiperOrigin-RevId: 488782173
 2022-11-15 16:07:17 -08:00
Camillo Lugaresi
a4fe3eb094 Add CreateBufferWithoutPool method to base pools 
This may not fit exactly in a pool class, but it makes it easy for the multi-pool to find the appropriate method by depending only on the type of the base pool. For the CVPixelBuffer case, the buffer type is CFHolder<CVPixelBufferRef>, and it seems even less appropriate to specialize that template to add such a method there. An alternative would be to allow defining a creation function separately.

PiperOrigin-RevId: 488782054
 2022-11-15 16:05:53 -08:00
Camillo Lugaresi
0d273dd11a Factor out CvPixelBufferPoolWrapper 
This is platform-specific and does not need to live in the main multi_pool sources.

PiperOrigin-RevId: 488781934
 2022-11-15 16:04:27 -08:00
Camillo Lugaresi
ab2dd779e7 Factor out CvTextureCacheManager 
This is a platform-specific component that is only used with CVPixelBufferPool.

PiperOrigin-RevId: 488781757
 2022-11-15 16:02:53 -08:00
MediaPipe Team
fae55910f4 Enable absl::string_view kCalculatorName 
PiperOrigin-RevId: 488781493
 2022-11-15 15:59:09 -08:00
Camillo Lugaresi
a520d6cc38 Remove FORCE_CONTIGUOUS_PIXEL_BUFFER_ON_IPHONE_SIMULATOR 
This workaround code is no longer necessary, as per the comment.

PiperOrigin-RevId: 488777606
 2022-11-15 15:41:53 -08:00
Camillo Lugaresi
3c71c64be1 Remove shared_ptr from SimplePool definition 
This makes the types more explicit and will help with factoring out platform-specific code.

PiperOrigin-RevId: 488775470
 2022-11-15 15:32:57 -08:00
Camillo Lugaresi
a67069156e Use flat_hash_map in ResourceCache 
This is the recommended hashmap in most cases.

PiperOrigin-RevId: 488772031
 2022-11-15 15:23:31 -08:00
MediaPipe Team
38b636f7ee Internal change 
PiperOrigin-RevId: 488770794
 2022-11-15 15:19:48 -08:00
Camillo Lugaresi
7a87546c30 Internal change 
PiperOrigin-RevId: 488761646
 2022-11-15 14:39:58 -08:00
MediaPipe Team
e65f21e2d8 Update the docstring to make it consistent with the model option update. 
PiperOrigin-RevId: 488761331
 2022-11-15 14:36:48 -08:00
Sebastian Schmidt
496720308c Migrate remaining MP Tasks Libraries to ts_declarations 
PiperOrigin-RevId: 488752799
 2022-11-15 14:08:15 -08:00
MediaPipe Team
1689112b23 Improve model_util_test code. 
PiperOrigin-RevId: 488752497
 2022-11-15 14:04:39 -08:00
MediaPipe Team
f14645cb06 Model maker gesture recognizer test changes 
PiperOrigin-RevId: 488702055
 2022-11-15 10:50:32 -08:00
MediaPipe Team
ebba119f15 Add Java ImageEmbedder API. 
PiperOrigin-RevId: 488588010
 2022-11-15 01:26:00 -08:00
Camillo Lugaresi
6f54308c25 Internal change 
PiperOrigin-RevId: 488552135
 2022-11-14 21:42:41 -08:00
Jiuqiang Tang
87dff8142c Fix a typo. 
PiperOrigin-RevId: 488519113
 2022-11-14 18:13:50 -08:00
Sebastian Schmidt
9bd8b432c3 Add typings support to d.ts package 
PiperOrigin-RevId: 488519074
 2022-11-14 18:10:25 -08:00
Jiuqiang Tang
f16e63694e Build embedding tasks into tasks AARs. 
PiperOrigin-RevId: 488509942
 2022-11-14 17:18:12 -08:00
Sebastian Schmidt
a12bc3fd0e Add IIFE bundles for NPM packages 
PiperOrigin-RevId: 488504360
 2022-11-14 16:51:58 -08:00
Jiuqiang Tang
794f64db55 Fix the wrong path of "text_embedder_graph_options_java_proto_lite". 
PiperOrigin-RevId: 488490050
 2022-11-14 15:51:05 -08:00
Jiuqiang Tang
cce1751dbf Expose tasks components API in PyPI packages. 
PiperOrigin-RevId: 488487666
 2022-11-14 15:41:11 -08:00
Jiuqiang Tang
e714e656fe Update python documentation. 
PiperOrigin-RevId: 488482142
 2022-11-14 15:18:47 -08:00
Sebastian Schmidt
c027373688 Internal change 
PiperOrigin-RevId: 488481286
 2022-11-14 15:15:22 -08:00
Sebastian Schmidt
11270d0c93 Image Embedder for Web 
PiperOrigin-RevId: 488468214
 2022-11-14 14:26:15 -08:00
Jiuqiang Tang
b4fba6fe61 MediaPipe Tasks AudioEmbedder Java API 
PiperOrigin-RevId: 488456442
 2022-11-14 13:42:55 -08:00
Sebastian Schmidt
ca7b5e9d8b Fix Script loading 
PiperOrigin-RevId: 488440736
 2022-11-14 12:40:41 -08:00
Copybara-Service
9a2af2f2a1 Merge pull request #3853 from kinaryml:audio-embedder-python 
PiperOrigin-RevId: 488434586
 2022-11-14 12:16:45 -08:00
Sebastian Schmidt
b00236e86e Hand Landmarker Web API 
PiperOrigin-RevId: 488434079
 2022-11-14 12:13:27 -08:00
MediaPipe Team
34daba4747 Add Java TextEmbedder API. 
PiperOrigin-RevId: 488427327
 2022-11-14 11:49:56 -08:00
Jiuqiang Tang
b40b2ade14 Fix typos. 
PiperOrigin-RevId: 488416345
 2022-11-14 11:07:59 -08:00
Jiuqiang Tang
4b5c3521af Dividing the timestamp by 1000 when returning a "none" result object from GestureRecognizer and HandLandmarker APIs. 
PiperOrigin-RevId: 488409077
 2022-11-14 10:44:20 -08:00
MediaPipe Team
c7d531ebb2 AddTarget -> ConnectTo and documentation 
PiperOrigin-RevId: 488407930
 2022-11-14 10:40:06 -08:00
MediaPipe Team
05cb40ff79 MediaPipe TextEmbedder task for web 
PiperOrigin-RevId: 488373613
 2022-11-14 08:34:09 -08:00
MediaPipe Team
badaccfb04 Internal change 
PiperOrigin-RevId: 488333493
 2022-11-14 05:05:03 -08:00
Nikolay Chirkov
bc6240e989 Zero-initialize id<MTLDevice> etc. members in Tensor 
PiperOrigin-RevId: 488308585
 2022-11-14 02:31:15 -08:00
Sebastian Schmidt
6610ca72ba Add ts_declaration rule for OSS 
PiperOrigin-RevId: 488307893
 2022-11-14 02:27:21 -08:00
Sebastian Schmidt
8d9cdb86dc Internal change 
PiperOrigin-RevId: 488294157
 2022-11-14 01:25:57 -08:00
kinaryml
a2bb58382d Added AudioEmbedder API and tests along with fixing a couple of typos in AudioClassifier 2022-11-14 01:20:45 -08:00
Jiuqiang Tang
6c0ca947de MediaPipe Tasks Audio embedder C++ API. 
PiperOrigin-RevId: 488273381
 2022-11-13 23:10:23 -08:00
Copybara-Service
0dfa91a166 Merge pull request #3846 from kinaryml:text-embedder-python 
PiperOrigin-RevId: 488198025
 2022-11-13 10:44:54 -08:00
Jiuqiang Tang
8aa59faaa4 Internal change 
PiperOrigin-RevId: 488193046
 2022-11-13 09:51:48 -08:00
Alan Kelly
bb6fa85703 Enable unsigned quantized infererence using XNNPACK. 
PiperOrigin-RevId: 488179656
 2022-11-13 07:10:04 -08:00
Liam Miller-Cushon
da36468409 Internal change 
PiperOrigin-RevId: 488065083
 2022-11-12 10:13:58 -08:00
MediaPipe Team
aafbf73c0a Renames model options in TextClassifier. 
PiperOrigin-RevId: 488063797
 2022-11-12 10:02:19 -08:00
kinaryml
a7ed160a8e Fixed a bug in embedding_result 2022-11-12 08:55:56 -08:00
kinaryml
a3788a23bc Removed unused code in image_embedder_test 2022-11-12 07:48:55 -08:00
kinaryml
157092d93e Removed unused dataclasses 2022-11-12 07:47:32 -08:00
kinaryml
a8103629c7 Updated Text Embedder API 2022-11-12 07:42:46 -08:00
Kinar R
acd9c280c0
Merge branch 'master' into text-embedder-python 2022-11-12 21:08:38 +05:30
Jiuqiang Tang
bf6c8a0b63 Expose ImageEmbedder APIs in PyPI packages. 
PiperOrigin-RevId: 488033833
 2022-11-12 04:35:22 -08:00
Copybara-Service
5d9ea88815 Merge pull request #3845 from kinaryml:image-embedder-python 
PiperOrigin-RevId: 487950862
 2022-11-11 17:03:00 -08:00
MediaPipe Team
c2ac040a6c Adds a public import API for TextClassifier. 
PiperOrigin-RevId: 487949023
 2022-11-11 16:52:16 -08:00
MediaPipe Team
8ec83d2aa0 Clarify AssetManager usage 
PiperOrigin-RevId: 487935478
 2022-11-11 15:49:50 -08:00
Sebastian Schmidt
c7030ac7fa Use CommonJS for NPM package 
PiperOrigin-RevId: 487902199
 2022-11-11 13:25:25 -08:00
Sebastian Schmidt
20a6f15f18 Improvements to NPM package 
PiperOrigin-RevId: 487901715
 2022-11-11 13:22:01 -08:00
Kinar R
fae77fc742
Update text_embedder_test.py 2022-11-12 01:27:20 +05:30
MediaPipe Team
340d7651af Internal change 
PiperOrigin-RevId: 487881149
 2022-11-11 11:53:41 -08:00
Hadon Nash
a83d87e157 Internal change 
PiperOrigin-RevId: 487880137
 2022-11-11 11:50:25 -08:00
Jiuqiang Tang
ce292c2a49 Fix a typo. 
PiperOrigin-RevId: 487872120
 2022-11-11 11:20:14 -08:00
MediaPipe Team
b0583a1821 Fixes SetAlphaCalculator silently failing to convert float inputs. 
PiperOrigin-RevId: 487868409
 2022-11-11 11:06:26 -08:00
Jiuqiang Tang
f329e38dc1 Map the "com_github_glog_glog" dependency of "com_google_sentencepiece" to mediapipe's "com_github_glog_glog_no_gflags". 
PiperOrigin-RevId: 487727239
 2022-11-10 22:12:42 -08:00
Kinar R
2d0b237711
Merge branch 'google:master' into text-embedder-python 2022-11-11 10:39:57 +05:30
Kinar R
f27068c6f2
Merge branch 'google:master' into image-embedder-python 2022-11-11 10:38:52 +05:30
MediaPipe Team
b4972ed4ae Update GestureRecognitionResult to GestureRecognizerResult. 
PiperOrigin-RevId: 487712873
 2022-11-10 20:25:31 -08:00
MediaPipe Team
3e05871f98 Open source Model Maker text tasks. 
PiperOrigin-RevId: 487706929
 2022-11-10 19:52:51 -08:00
Mark McDonald
8c8b668ba3 Fix comment for mediapipe license. 
The `"""` comment indicates a public docstring for the module, and will end up in the generated docs. By using a "private" comment (`#`) we will not document the license as part of the API.

The Apache license is noted in the footer of generated docs, and this is sufficient.

PiperOrigin-RevId: 487688719
 2022-11-10 18:06:35 -08:00
MediaPipe Team
ec327cedcb Internal Changes 
PiperOrigin-RevId: 487673720
 2022-11-10 16:48:13 -08:00
MediaPipe Team
2ea5184c51 Internal change 
PiperOrigin-RevId: 487663454
 2022-11-10 16:04:22 -08:00
Sebastian Schmidt
1a59f0a217 Prevent renaming of properties used in API 
PiperOrigin-RevId: 487661625
 2022-11-10 15:58:47 -08:00
Jiuqiang Tang
b3ac1c1387 Changes to create MediaPipe Tasks Audio AAR. 
PiperOrigin-RevId: 487661296
 2022-11-10 15:55:33 -08:00
Jiuqiang Tang
d6ba5e4bbe Map the "com_github_glog_glog" dependency of "com_google_audio_tools" to mediapipe's "com_github_glog_glog_no_gflags". 
PiperOrigin-RevId: 487651038
 2022-11-10 15:11:57 -08:00
Copybara-Service
bbf4ff0300 Merge pull request #3832 from kinaryml:update-classification-tasks-output-stream 
PiperOrigin-RevId: 487638077
 2022-11-10 14:22:31 -08:00
Hadon Nash
d93caf00de Fixing a bug in the BypassCalculator "passthrough" output stream ids. 
PiperOrigin-RevId: 487633708
 2022-11-10 14:05:47 -08:00
MediaPipe Team
8ec4427bd7 Add option, result types and utils for Java embedders. 
PiperOrigin-RevId: 487615327
 2022-11-10 12:53:00 -08:00
Sebastian Schmidt
aeb2466844 Pick TFLite delegate for Web CPU 
PiperOrigin-RevId: 487611446
 2022-11-10 12:37:42 -08:00
Jiuqiang Tang
bb5782ee2f Correct the timestamp in the vision tasks result objects. 
PiperOrigin-RevId: 487597999
 2022-11-10 11:46:16 -08:00
Jiuqiang Tang
da4d455d0c Implement MediaPipe Tasks AudioClassifier Java API. 
PiperOrigin-RevId: 487570148
 2022-11-10 10:10:03 -08:00
Jiuqiang Tang
2ce3a9719e Remove the problematic packet release calls in OutputHandler::run. 
OutputHandler::run is installed as a MultiStreamCallback into the TaskRunner and it's invoked in the Graph::CallbackToJava() jni method.

In CallbackToJava(), the packets are removed in RemovePacket() calls after the multi stream callback is invoked. The current logic lets the MultiStreamCallback release the packets first and leaves null pointers to RemovePacket() to remove, which may cause the app crash. The fix is to remove the packet release calls in the MultiStreamCallback and always leave the packet release work to RemovePacket().

PiperOrigin-RevId: 487558809
 2022-11-10 09:27:35 -08:00
Kinar R
0a6e21c212
Merge branch 'master' into image-embedder-python 2022-11-10 16:04:28 +05:30
kinaryml
0e9b925726 Fixed some typos and revised image embedder tests 2022-11-10 02:30:17 -08:00
kinaryml
1604908a59 Added files needed for the text embedder's implementation and tests 2022-11-10 02:16:51 -08:00
MediaPipe Team
0b12aa9435 Internal change 
PiperOrigin-RevId: 487466061
 2022-11-10 01:32:14 -08:00
MediaPipe Team
0ac604d507 Add timestamp aggregation to EmbeddingPostprocessingGraph. 
PiperOrigin-RevId: 487463848
 2022-11-10 01:20:45 -08:00
MediaPipe Team
f11c757629 Update C++ HandLandmarksDetectionResult to HandLandmarkerResult. 
PiperOrigin-RevId: 487443827
 2022-11-09 23:20:15 -08:00
Yuqi Li
d2142e86a9 Metadata Writer: Add Bert metadata writer in Text Classifier. 
PiperOrigin-RevId: 487354439
 2022-11-09 14:53:12 -08:00
MediaPipe Team
116b4bb6c4 Internal change 
PiperOrigin-RevId: 487338991
 2022-11-09 13:57:13 -08:00
MediaPipe Team
7066ee231e Hand Landmarker Java API and unit test. 
PiperOrigin-RevId: 487307858
 2022-11-09 11:55:31 -08:00
kinaryml
7ec0d8cf3b Added tolerance for vector coordinate values in embeddings 2022-11-09 11:53:30 -08:00
Jiuqiang Tang
6f66a7fde4 Rename GestureRecognitionResult to GestureRecognizerResult. 
PiperOrigin-RevId: 487306292
 2022-11-09 11:49:29 -08:00
kinaryml
17bb174444 Refactored embeddings to embedding_result 2022-11-09 11:42:32 -08:00
Jiuqiang Tang
4e568a1c3d Expose HandLandmarker API in PyPI packages. 
PiperOrigin-RevId: 487294501
 2022-11-09 11:06:33 -08:00
Sebastian Schmidt
84895316d2 Add pre-compiled WASM files to NPM packages 
PiperOrigin-RevId: 487283112
 2022-11-09 10:28:09 -08:00
Copybara-Service
0a5534204f Merge pull request #3829 from kinaryml:hand-landmarker-python 
PiperOrigin-RevId: 487281485
 2022-11-09 10:23:38 -08:00
kinaryml
dc30cf9732 Updated embeddings container to use a timestamp and made parameters - head_index,head_name optional 2022-11-08 23:51:55 -08:00
kinaryml
d327b3649d Fixed some typos in methods and comments 2022-11-08 23:47:58 -08:00
kinaryml
dd6fdedd5f Added some sanity tests 2022-11-08 23:15:58 -08:00
Kinar R
36c50ff8f3
Merge branch 'google:master' into image-embedder-python 2022-11-09 12:41:55 +05:30
kinaryml
7e251bc6b6 Merge branch 'hand-landmarker-python' of https://github.com/kinaryml/mediapipe into hand-landmarker-python 2022-11-08 23:07:30 -08:00
Kinar R
812fa2cc70
Merge branch 'google:master' into hand-landmarker-python 2022-11-09 12:37:06 +05:30
kinaryml
46f135e54d Renamed HandLandmarksDetectionResult to HandLandmarkerResult 2022-11-08 23:06:28 -08:00
MediaPipe Team
b4e1833dd0 Internal change 
PiperOrigin-RevId: 487125366
 2022-11-08 20:35:46 -08:00
MediaPipe Team
a5bcb97d88 Adds an AverageWordVecModel. 
PiperOrigin-RevId: 487104909
 2022-11-08 18:29:21 -08:00
MediaPipe Team
c31aaa94a6 Adds a BertClassifier. 
PiperOrigin-RevId: 487086744
 2022-11-08 16:49:46 -08:00
Sebastian Schmidt
669d539551 NPM package definitions for MediaPipe Tasks 
PiperOrigin-RevId: 487071334
 2022-11-08 15:46:05 -08:00
Lu Wang
253ff0f85c Update the Java doc for model asset for BaseOptions 
PiperOrigin-RevId: 487059371
 2022-11-08 14:57:27 -08:00
MediaPipe Team
0917e8cb8e Support continual training image classifier from saved checkpoint files. 
PiperOrigin-RevId: 487057612
 2022-11-08 14:50:04 -08:00
MediaPipe Team
b3d19fa1af Use model bundle writer when exporting models in gesture recognizer 
PiperOrigin-RevId: 487042776
 2022-11-08 13:52:08 -08:00
MediaPipe Team
0363d60511 Open-sources TextEmbedder. 
PiperOrigin-RevId: 487041832
 2022-11-08 13:48:49 -08:00
Kinar R
37930609ff
Update landmark_detection_result.py 2022-11-09 02:58:38 +05:30
MediaPipe Team
ace098f370 Add proper Cast for MultiPort 
PiperOrigin-RevId: 487012509
 2022-11-08 11:57:03 -08:00
kinaryml
96e2eb38c7 Merge branch 'hand-landmarker-python' of https://github.com/kinaryml/mediapipe into hand-landmarker-python 2022-11-08 11:27:13 -08:00
kinaryml
77cc4c53ab Updated the 'test_create_from_options_fails_with_invalid_model_path' test case 2022-11-08 11:26:02 -08:00
Kinar R
6dd6d8921f
Merge branch 'google:master' into hand-landmarker-python 2022-11-09 00:52:20 +05:30
MediaPipe Team
7a1e55b872 Adds unit tests for TextEmbedder. 
PiperOrigin-RevId: 486997035
 2022-11-08 11:01:04 -08:00
kinaryml
72f08c6878 Used loops instead of list comprehension 2022-11-08 10:57:12 -08:00
kinaryml
647b94240d Updated docstring 2022-11-08 10:54:44 -08:00
Jiuqiang Tang
ff6ca706c6 Internal change 
PiperOrigin-RevId: 486994334
 2022-11-08 10:52:51 -08:00
kinaryml
88ab9d4922 Used loops instead of complex list comprehensions 2022-11-08 10:52:26 -08:00
kinaryml
c0265ca2ac Fixed some typos 2022-11-08 10:44:21 -08:00
kinaryml
e8d771baf3 Updated landmark_detection_result 2022-11-08 10:37:37 -08:00
Jiuqiang Tang
5c8b95666a Update "test_create_from_options_fails_with_invalid_model_path" of all tasks python api tests. 
PiperOrigin-RevId: 486990017
 2022-11-08 10:36:24 -08:00
kinaryml
0fd78c2ec6 Updated classifier python tasks to use the new classification_result dataclass 2022-11-08 10:28:24 -08:00
MediaPipe Team
5e6842aa5c Migrate Java ImageClassifier / TextClassifier to new result format. 
PiperOrigin-RevId: 486976459
 2022-11-08 09:47:42 -08:00
Sebastian Schmidt
26066787b3 Allow Web developers to opt into CPU or GPU processing 
PiperOrigin-RevId: 486935157
 2022-11-08 06:37:38 -08:00
kinaryml
9c9519eeb8 Fixed a to_pb2() bug in the landmark_detection_result container 2022-11-08 02:55:19 -08:00
kinaryml
0402ee383f Added hand landmarker Python API and tests 2022-11-08 01:05:36 -08:00
MediaPipe Team
4a6562d423 internal 
PiperOrigin-RevId: 486853812
 2022-11-07 22:39:23 -08:00
MediaPipe Team
24d03451c7 Internal change - migration 
PiperOrigin-RevId: 486853689
 2022-11-07 22:36:22 -08:00
Yuqi Li
0a08e4768b Metadata Writer: Add metadata writer for Text Classifier. 
PiperOrigin-RevId: 486844428
 2022-11-07 21:25:44 -08:00
Jiuqiang Tang
b14178d305 Temporarily exclude "audio_classifier_graph" from "builtin_task_graphs" on Windows. 
PiperOrigin-RevId: 486806337
 2022-11-07 17:27:33 -08:00
MediaPipe Team
716c7e3b8a Internal change 
PiperOrigin-RevId: 486805777
 2022-11-07 17:23:46 -08:00
MediaPipe Team
c3bb4bb5da Refactor image classifier Create API by defining an ImageClassifierOption which wraps the model_spec, model_options, and hparams. Also migrate the definition of HParams by extending it from BaseHParams. 
PiperOrigin-RevId: 486796657
 2022-11-07 16:39:35 -08:00
MediaPipe Team
1049ef781d Add image_segmenter namespace and update image segmenter proto for java package. 
PiperOrigin-RevId: 486792164
 2022-11-07 16:24:16 -08:00
Camillo Lugaresi
1a151d9ffb Internal change 
PiperOrigin-RevId: 486792117
 2022-11-07 16:21:15 -08:00
MediaPipe Team
6f38a7a21f Add hand landmarker C++ api. 
PiperOrigin-RevId: 486791711
 2022-11-07 16:17:42 -08:00
Kinar R
83608d4670
Merge branch 'master' into image-embedder-python 2022-11-08 05:02:46 +05:30
Jiuqiang Tang
571c0b1fef Expose MediaPipe Tasks Audio and Text APIs in MediaPipe PyPI packages. 
PiperOrigin-RevId: 486777537
 2022-11-07 15:17:57 -08:00
Jiuqiang Tang
63a759accc Implement MediaPipe AudioClassifier Tasks Python API. Adjust the AudioClassifier Tasks C++ API to remove "sample_rate" from its options. 
PiperOrigin-RevId: 486763992
 2022-11-07 14:28:14 -08:00
MediaPipe Team
51dbd9779c Introduce Stream/SidePacket aliases for Source/SideSource 
PiperOrigin-RevId: 486761126
 2022-11-07 14:20:08 -08:00
Sebastian Schmidt
e994efd40f Add entry points for NPM packages 
PiperOrigin-RevId: 486760105
 2022-11-07 14:16:29 -08:00
kinaryml
664d9c49e7 Revised image embedder implementation 2022-11-07 13:59:07 -08:00
Kinar R
ba1ee5b404
Merge branch 'master' into image-embedder-python 2022-11-08 02:53:20 +05:30
MediaPipe Team
6b54cae34c Internal change - cleanup 
PiperOrigin-RevId: 486721059
 2022-11-07 11:42:50 -08:00
Sebastian Schmidt
2371051e17 Open Source the remaining Web implementation files 
PiperOrigin-RevId: 486701932
 2022-11-07 10:34:44 -08:00
Jiuqiang Tang
c02f3933d4 Temporarily exclude "text_classifier_graph" from "builtin_task_graphs" on Windows. 
PiperOrigin-RevId: 486700027
 2022-11-07 10:28:28 -08:00
Jiuqiang Tang
9f4496db77 Fix typos 
PiperOrigin-RevId: 486698923
 2022-11-07 10:24:17 -08:00
Jiuqiang Tang
17a2de8cf7 Rename TextClassificationResult to TextClassifierResult. 
PiperOrigin-RevId: 486685936
 2022-11-07 09:39:01 -08:00
Copybara-Service
4c06303ec7 Merge pull request #3820 from kinaryml:text-classifier-python 
PiperOrigin-RevId: 486566800
 2022-11-06 22:15:09 -08:00
MediaPipe Team
4b0423e3a2 Internal change 
PiperOrigin-RevId: 486547910
 2022-11-06 19:34:24 -08:00
Jiuqiang Tang
9504c5e6a1 Add a create_from_array classmethod to the AudioData class. 
PiperOrigin-RevId: 486310154
 2022-11-05 00:09:30 -07:00
kinaryml
b44c49250b Removed unused constant 2022-11-04 22:36:13 -07:00
Kinar R
740d2e47b5
Merge branch 'google:master' into text-classifier-python 2022-11-05 11:03:24 +05:30
Hadon Nash
91782a2772 Internal change 
PiperOrigin-RevId: 486283316
 2022-11-04 19:48:18 -07:00
Jiuqiang Tang
416f91180b Remove the use of designated initializers. 
PiperOrigin-RevId: 486261384
 2022-11-04 16:57:21 -07:00
Jiuqiang Tang
35f635d8ff Add an argument to packet_creator.create_matrix to allow the input matrix to be transposed first. 
PiperOrigin-RevId: 486258078
 2022-11-04 16:39:32 -07:00
Jiuqiang Tang
c10ebe8476 Remove the use of designated initializers. 
PiperOrigin-RevId: 486215798
 2022-11-04 13:24:48 -07:00
Kinar R
b1d1ef0124
Merge branch 'google:master' into image-embedder-python 2022-11-05 00:17:26 +05:30
MediaPipe Team
5e1a2fcdbb Update EmbeddingResult format and dependent tasks. 
PiperOrigin-RevId: 486186491
 2022-11-04 11:20:07 -07:00
Jiuqiang Tang
66e591d4bc Remove the use of designated initializers. 
PiperOrigin-RevId: 486185274
 2022-11-04 11:15:20 -07:00
Sebastian Schmidt
f05868d8e0 Set up the Open Source build for MediaPipe Tasks Web 
PiperOrigin-RevId: 486181061
 2022-11-04 10:59:32 -07:00
Jiuqiang Tang
5fd3701cfd Expose ImageSegmenter and GestureRecognizer APIs in PyPI packages. 
PiperOrigin-RevId: 486176206
 2022-11-04 10:39:21 -07:00
MediaPipe Team
8b2c937b9e Migrate AudioClassifier C++ to use new ClassificationResult struct. 
PiperOrigin-RevId: 486162683
 2022-11-04 09:46:09 -07:00
MediaPipe Team
62cd67e996 Internal change 
PiperOrigin-RevId: 486148328
 2022-11-04 08:42:39 -07:00
Jiuqiang Tang
5f5f50d8f7 Implement MediaPipe Tasks Python AudioData. 
PiperOrigin-RevId: 486147173
 2022-11-04 08:32:30 -07:00
Copybara-Service
5024c815f1 Merge pull request #3813 from kinaryml:image-segmenter-minor-fixes 
PiperOrigin-RevId: 486051659
 2022-11-03 21:41:37 -07:00
MediaPipe Team
b2cc2cc60c Internal changes 
PiperOrigin-RevId: 486014305
 2022-11-03 17:23:20 -07:00
MediaPipe Team
06cb73fc81 Modified internal dependencies. 
PiperOrigin-RevId: 485992876
 2022-11-03 15:44:36 -07:00
kinaryml
e2d50745ac Revised image embedder implementation 2022-11-03 14:30:21 -07:00
MediaPipe Team
d29c3d7512 Add metadata writer to image_classifier in model_maker 
PiperOrigin-RevId: 485926985
 2022-11-03 11:30:38 -07:00
kinaryml
492607152a Reverted changes to rect 2022-11-03 10:56:38 -07:00
Kinar R
5a68ba84b6
Merge branch 'master' into image-embedder-python 2022-11-03 23:24:31 +05:30
Jiuqiang Tang
b472d8ff66 Add the missing __init__.py file. 
PiperOrigin-RevId: 485892501
 2022-11-03 09:24:12 -07:00
Copybara-Service
716e59f90c Merge pull request #3801 from kinaryml:gesture-recognizer-python 
PiperOrigin-RevId: 485884796
 2022-11-03 08:52:17 -07:00
kinaryml
14e18a03d6 Removed unused custom classifier options in tests 2022-11-03 02:54:57 -07:00
kinaryml
de619e2702 Removed a comment 2022-11-03 02:33:50 -07:00
kinaryml
2001e2b77c Removed unused file 2022-11-03 02:32:55 -07:00
kinaryml
6e9a070dd1 Added text classifier implementation and tests 2022-11-03 02:29:40 -07:00
kinaryml
35a04522fa Moved the OutputType and Activation classes to ImageSegmenter's inner classes 2022-11-03 00:46:00 -07:00
MediaPipe Team
ddf37d014e Fix test utils bug when two file names have same ending. 
PiperOrigin-RevId: 485780917
 2022-11-02 22:33:36 -07:00
Jiuqiang Tang
ac6aa9ce10 Fix a bug in the AndroidPacketCreator's createImage method. 
PiperOrigin-RevId: 485716828
 2022-11-02 15:56:49 -07:00
Jiuqiang Tang
01d87f0867 Expose MediaPipe Tasks in MediaPipe PyPI packages. 
PiperOrigin-RevId: 485690163
 2022-11-02 14:07:47 -07:00
MediaPipe Team
0bf5110a7e Add a model_maker internal library directory. 
PiperOrigin-RevId: 485665201
 2022-11-02 12:31:58 -07:00
MediaPipe Team
d61ab92b90 Add classification result structs and use then in ImageClassifier & TextClassifier. 
PiperOrigin-RevId: 485567133
 2022-11-02 05:13:12 -07:00
MediaPipe Team
aab5f84aae Internal change 
PiperOrigin-RevId: 485553891
 2022-11-02 03:49:48 -07:00
MediaPipe Team
475e6b4fd5 Modify ClassificationAggregationCalculator to output new unified formats. 
PiperOrigin-RevId: 485546547
 2022-11-02 03:11:20 -07:00
MediaPipe Team
f1f123d255 Extracted common logics from the ImageToTensorCalculator such that it can be 
reused by other calculators.

PiperOrigin-RevId: 485472451
 2022-11-01 18:43:46 -07:00
MediaPipe Team
aaf98ea43c Add CPU invoke to Perfetto trace 
PiperOrigin-RevId: 485464221
 2022-11-01 17:50:39 -07:00
MediaPipe Team
ecd93ff080 Internal changes 
PiperOrigin-RevId: 485445138
 2022-11-01 16:18:06 -07:00
Kinar R
3a2f30185f
Merge branch 'master' into gesture-recognizer-python 2022-11-02 04:11:18 +05:30
kinaryml
c5765ac836 Refactored Rect to use top-left coordinates and appropriately updated the Image Classifier and Gesture Recognizer APIs/tests 2022-11-01 15:37:00 -07:00
Yuqi Li
e719f7b4e2 Model Maker Gesture Recognizer: add metadata writer and create model bundle. 
PiperOrigin-RevId: 485426865
 2022-11-01 15:02:45 -07:00
MediaPipe Team
c6a64683f6 Add ConcatenateClassificationListCalculator 
PiperOrigin-RevId: 485398597
 2022-11-01 13:13:48 -07:00
MediaPipe Team
700971de70 Internal change 
PiperOrigin-RevId: 485396135
 2022-11-01 13:05:43 -07:00
Copybara-Service
6e0397b226 Merge pull request #3800 from kinaryml:python-test-proto-equals 
PiperOrigin-RevId: 485340924
 2022-11-01 09:42:19 -07:00
Frederic Rechtenstein
60a19f4fe2 Fix missing space in docstring. 
PiperOrigin-RevId: 485337309
 2022-11-01 09:27:29 -07:00
Jiuqiang Tang
e4899c8207 MediaPipe Python audio base layer. 
PiperOrigin-RevId: 485329470
 2022-11-01 08:53:36 -07:00
kinaryml
a913255080 Removed min score thres from tests 2022-10-31 23:07:05 -07:00
kinaryml
d3b472e888 Add allow_list/deny_list support 2022-10-31 22:16:37 -07:00
Kinar R
1aaaca1e12
Merge branch 'google:master' into gesture-recognizer-python 2022-11-01 08:16:43 +05:30
Jiuqiang Tang
86951e048d Remove the unused TaskRunner from the mediapipe tasks python code. 
PiperOrigin-RevId: 485181210
 2022-10-31 16:54:30 -07:00
MediaPipe Team
9f8b5e5c11 Add allow_list/deny_list to gesture recognizer options. 
PiperOrigin-RevId: 485141209
 2022-10-31 13:57:06 -07:00
Jiuqiang Tang
4717ac298c Add EndLoopMatrixCalculator 
PiperOrigin-RevId: 485097386
 2022-10-31 11:08:21 -07:00
kinaryml
2b5a077579 Updated comments 2022-10-31 05:48:45 -07:00
kinaryml
d635b4281e Added a test for the canned classification of the gesture victory 2022-10-31 05:47:28 -07:00
kinaryml
888ddd4b74 Removed unused classifier options proto 2022-10-31 05:37:24 -07:00
kinaryml
19be9e9012 Revised gesture recognizer implementation 2022-10-31 05:34:31 -07:00
Kinar R
5ec87c8bd2
Merge branch 'master' into gesture-recognizer-python 2022-10-31 16:47:43 +05:30
MediaPipe Team
459214e6a3 Rename index_by_label to label_names. 
PiperOrigin-RevId: 484956259
 2022-10-30 22:26:17 -07:00
Copybara-Service
7bcf322625 Merge pull request #3739 from kinaryml:image-segmenter-python-impl 
PiperOrigin-RevId: 484922757
 2022-10-30 17:11:42 -07:00
kinaryml
fb4872b068 Refactored code and removed some issues 2022-10-30 15:42:26 -07:00
MediaPipe Team
be13a76928 Integrate canned and custom gesture classifier. 
PiperOrigin-RevId: 484914164
 2022-10-30 15:36:09 -07:00
kinaryml
4b66599419 Updated docstring in gesture_recognizer 2022-10-30 09:10:15 -07:00
Kinar R
6f485ae3dd
Merge branch 'google:master' into gesture-recognizer-python 2022-10-30 20:55:53 +05:30
kinaryml
f62cfd1690 Removed classification proto to use the existing category dataclass instead and removed NormalizedLandmarkList and LandmarkList dataclasses 2022-10-30 08:23:14 -07:00
Hadon Nash
313b387ced Adding SwitchContainer option "tick_input_stream" to specify which input streams indicate the output timestamps to expect from the contained graphs. 
PiperOrigin-RevId: 484615286
 2022-10-28 14:04:20 -07:00
Jiuqiang Tang
e16be2e8fa Implement the initial version of TensorsToAudioCalculator that supports ifft and inverse hann windowing. 
PiperOrigin-RevId: 484605092
 2022-10-28 13:19:43 -07:00
Hadon Nash
c5c639d634 Internal change 
PiperOrigin-RevId: 484583911
 2022-10-28 11:46:10 -07:00
Jiuqiang Tang
f4f8b11ffc Fix the comment. 
PiperOrigin-RevId: 484571254
 2022-10-28 10:57:13 -07:00
Kinar R
334f641463
Merge branch 'master' into image-segmenter-python-impl 2022-10-28 22:50:33 +05:30
Kinar R
0f7c5d5e90
Merge branch 'google:master' into gesture-recognizer-python 2022-10-28 14:08:52 +05:30
kinaryml
b81b5a9035 Added a test for min_gesture_confidence 2022-10-28 01:38:15 -07:00
MediaPipe Team
de5fe27e05 Modified internal dependencies. 
PiperOrigin-RevId: 484407262
 2022-10-27 19:08:19 -07:00
MediaPipe Team
abd1ff66c8 Fix https://github.com/google/mediapipe/issues/3784 
PiperOrigin-RevId: 484365654
 2022-10-27 15:32:44 -07:00
MediaPipe Team
fc1d75cc99 Add CombinedPredictionCalculator. 
PiperOrigin-RevId: 484301880
 2022-10-27 11:16:12 -07:00
MediaPipe Team
ee84e447b2 Internal change 
PiperOrigin-RevId: 484299808
 2022-10-27 11:08:31 -07:00
kinaryml
8194513934 Updated docstrings and removed the redundant private helper method 2022-10-26 23:56:54 -07:00
Jiuqiang Tang
a941c5cdd1 Create MediaPipe "tasks-text" AAR. 
PiperOrigin-RevId: 484004494
 2022-10-26 10:18:22 -07:00
MediaPipe Team
254f786624 Add an option to set image preprocessing backend as gpu. 
PiperOrigin-RevId: 483888202
 2022-10-26 00:46:19 -07:00
kinaryml
f8a98ccba4 Added a test utility method to compare protos directly 2022-10-25 23:38:32 -07:00
kinaryml
fbf7ba6f1a Reverted some changes to rect 2022-10-25 23:15:16 -07:00
kinaryml
75af46d273 Revised API to align with recent changes 2022-10-25 23:13:12 -07:00
Kinar R
0de97497fa
Merge branch 'master' into gesture-recognizer-python 2022-10-26 11:37:12 +05:30
Copybara-Service
ae5b09e2b2 Merge pull request #3738 from kinaryml:image-classification-python-impl 
PiperOrigin-RevId: 483818404
 2022-10-25 17:26:32 -07:00
MediaPipe Team
a28c9d2c26 Set steps_per_epoch to None when calling model.fit() method for image classifier. 
PiperOrigin-RevId: 483764377
 2022-10-25 13:44:03 -07:00
MediaPipe Team
36bd9abb8f Internal change 
PiperOrigin-RevId: 483751427
 2022-10-25 12:53:39 -07:00
MediaPipe Team
21abfc9125 Update gpu origin. 
PiperOrigin-RevId: 483742652
 2022-10-25 12:14:28 -07:00
kinaryml
8762d15c81 Added remaining tests for the GestureRecognizer Python MediaPipe Tasks API 2022-10-25 11:11:15 -07:00
MediaPipe Team
d240c009e2 Remove unnecessary location_data_proto dependency on rect_proto. 
PiperOrigin-RevId: 483679555
 2022-10-25 08:24:07 -07:00
kinaryml
18eb089d39 Added a simple test to verify gesture recognition results 2022-10-25 07:38:04 -07:00
MediaPipe Team
6b0a7fb281 Reverting back to special handling for Egl Thread Exit on Android 
PiperOrigin-RevId: 483505151
 2022-10-24 15:45:35 -07:00
MediaPipe Team
94cd134809 Add support for image rotation in Java vision tasks. 
PiperOrigin-RevId: 483493729
 2022-10-24 15:01:23 -07:00
MediaPipe Team
3d88b1797a Switch CHECK to a status in resource handling code. Expand error message. 
PiperOrigin-RevId: 483438131
 2022-10-24 11:27:15 -07:00
Jiuqiang Tang
2cf9523468 Fix the java_package name. 
PiperOrigin-RevId: 483428848
 2022-10-24 10:55:16 -07:00
MediaPipe Team
2f2baeff68 Add support for rotation in ImageEmbedder & ImageSegmenter C++ APIs 
PiperOrigin-RevId: 483416498
 2022-10-24 10:14:05 -07:00
MediaPipe Team
0fd69e8d83 Open-source some tokenizer unit tests. 
PiperOrigin-RevId: 483399326
 2022-10-24 09:12:52 -07:00
kinaryml
9a1a9d4c13 Added files needed for the GestureRecognizer API implementation 2022-10-24 06:08:27 -07:00
MediaPipe Team
af051dcb62 internal change 
PiperOrigin-RevId: 483308781
 2022-10-24 01:39:23 -07:00
Jiuqiang Tang
de93d06f87 Implement build rules and targets to create MediaPipe Tasks AARs 
PiperOrigin-RevId: 483297891
 2022-10-24 00:45:32 -07:00
Yuqi Li
ab17be9294 Metadata Writer: Add Metadata Writer for image classifier. 
PiperOrigin-RevId: 483282627
 2022-10-23 23:05:32 -07:00
MediaPipe Team
ec2a34d2a4 Replace pytype_struct_contrib_test by py_strict_test. Also remove unnecessary BUILD attributes. 
PiperOrigin-RevId: 483237371
 2022-10-23 16:45:02 -07:00
MediaPipe Team
ab0193bbb7 Internal change 
PiperOrigin-RevId: 483078695
 2022-10-22 15:55:46 -07:00
kinaryml
ebb2686fb4 Moved OutputType and Activation to image_segmenter 2022-10-22 03:34:26 -07:00
MediaPipe Team
d8006a2f87 Use model bundle for gesture recognizer. 
PiperOrigin-RevId: 482960305
 2022-10-21 21:54:34 -07:00
MediaPipe Team
404323f631 Add Mediapipe Tasks Gesture Recognizer benchmarks 
PiperOrigin-RevId: 482935780
 2022-10-21 18:33:49 -07:00
Jiuqiang Tang
abed54ea30 Rename the mediapipe java image container from Image to MPImage. 
PiperOrigin-RevId: 482933122
 2022-10-21 18:17:35 -07:00
MediaPipe Team
7196db275e Internal change 
PiperOrigin-RevId: 482925717
 2022-10-21 17:28:49 -07:00
MediaPipe Team
ea1d85d811 Update model_task_graph to support multiple local model resources. 
PiperOrigin-RevId: 482917453
 2022-10-21 16:42:58 -07:00
MediaPipe Team
4a6c23a76a Internal change 
PiperOrigin-RevId: 482906478
 2022-10-21 15:48:54 -07:00
MediaPipe Team
d0437b7f91 Add tensor_index and tensor_name fields to ClassificationList 
PiperOrigin-RevId: 482901854
 2022-10-21 15:27:35 -07:00
kinaryml
024a6866a7 Removed some unneeded aliases 2022-10-21 13:39:59 -07:00
MediaPipe Team
36d69971a7 Internal change 
PiperOrigin-RevId: 482875698
 2022-10-21 13:36:41 -07:00
Kinar R
1748663a5a
Merge branch 'google:master' into image-segmenter-python-impl 2022-10-22 02:04:57 +05:30
kinaryml
5231a0ad9f Removed SegmenterOptions dataclasses to enumerate options within ImageSegmenterOptions instead 2022-10-21 13:34:30 -07:00
Jiuqiang Tang
e1a9615e1a Add "alwayslink = 1" to "universal_sentence_encoder_preprocessor_calculator". 
PiperOrigin-RevId: 482864252
 2022-10-21 12:48:36 -07:00
MediaPipe Team
e453c3b206 Migrates image_preprocessing from core to vision. 
PiperOrigin-RevId: 482856336
 2022-10-21 12:16:54 -07:00
Sebastian Schmidt
55a536e55e Typo fix in Landmark 
PiperOrigin-RevId: 482854764
 2022-10-21 12:13:12 -07:00
Sebastian Schmidt
fe77c5489f Fix comments in GestureRecognizer 
PiperOrigin-RevId: 482854253
 2022-10-21 12:09:36 -07:00
Jiuqiang Tang
ed938a9214 Replace the usage of InvalidProtocolBufferException with IOException. 
PiperOrigin-RevId: 482844446
 2022-10-21 11:30:28 -07:00
MediaPipe Team
55ba23ce9a Cleans up unused function from image_classifier's Dataset. 
PiperOrigin-RevId: 482703775
 2022-10-20 23:59:20 -07:00
MediaPipe Team
348c4e6652 Add a test to test saving and loading the gesture recognizer model. 
Also update the loss function init method to resolve a bug when loading model.

PiperOrigin-RevId: 482702020
 2022-10-20 23:45:26 -07:00
MediaPipe Team
086fc442fd Internal change 
PiperOrigin-RevId: 482697999
 2022-10-20 23:15:17 -07:00
MediaPipe Team
9e2a9bb4be Internal change 
PiperOrigin-RevId: 482565462
 2022-10-20 12:33:30 -07:00
MediaPipe Team
4b5df1cb96 Add support for rotations in GestureRecognizer C++ API. 
PiperOrigin-RevId: 482533599
 2022-10-20 10:42:42 -07:00
MediaPipe Team
e71638cf67 Add an export_tflite API to gesture recognizer model maker library. 
PiperOrigin-RevId: 482527017
 2022-10-20 10:21:34 -07:00
kinaryml
71d5b69544 Added files for the image embedder implementation and a simple test 2022-10-20 02:29:14 -07:00
MediaPipe Team
467cd34feb Internal change 
PiperOrigin-RevId: 482401089
 2022-10-19 23:49:51 -07:00
MediaPipe Team
41d6f6d005 Adds a Java API for TextClassifier. 
PiperOrigin-RevId: 482394706
 2022-10-19 23:05:34 -07:00
MediaPipe Team
3d588bae8b Internal change 
PiperOrigin-RevId: 482259130
 2022-10-19 11:34:32 -07:00
MediaPipe Team
c260074abb Rename "Bound" struct to "Rect" and remove unused "Landmark" struct. 
PiperOrigin-RevId: 482255889
 2022-10-19 11:22:56 -07:00
MediaPipe Team
8426428de3 Internal cleanup for text-domain unit tests. 
PiperOrigin-RevId: 482241566
 2022-10-19 10:31:21 -07:00
MediaPipe Team
a18f91e04f Open-sources UniversalSentenceEncoderPreprocessorCalculator. 
PiperOrigin-RevId: 482222697
 2022-10-19 09:20:38 -07:00
MediaPipe Team
70df9e2419 Open-source the TextClassifier C++ API. 
PiperOrigin-RevId: 482218721
 2022-10-19 09:03:29 -07:00
MediaPipe Team
7a6ae97a0e Internal change 
PiperOrigin-RevId: 482135978
 2022-10-19 01:23:58 -07:00
MediaPipe Team
cfd5a20618 Open-sources the TextPreprocessingSubgraph. 
PiperOrigin-RevId: 482048438
 2022-10-18 16:14:33 -07:00
Sebastian Schmidt
0b55fbda34 Fix compilation error on Windows 
PiperOrigin-RevId: 482037921
 2022-10-18 15:31:22 -07:00
Hadon Nash
7785603fbe Adding BypassCalculator for use with SwitchContainer. 
PiperOrigin-RevId: 482030395
 2022-10-18 14:59:25 -07:00
MediaPipe Team
e86cd39521 Use unused pixel format attribute variable as intended 
Based on the comments and the context, we should be trying to use the specified attributes in orders to initialize the pixel_format_ variable.

PiperOrigin-RevId: 482011065
 2022-10-18 13:51:24 -07:00
MediaPipe Team
43345160c5 Add ImageFrame and GlTexture view providers to Ahwb GpuBuffer storage 
- ImageFrame views are backed by CPU locks of the AHWB.
- GlTextureViews are managed through a GlTextureBuffer connected to the AHWB allocation.

This improves the latency of heterogeneous access sequences, but makes `ImageFrame > ImageFrame` sequences slower due to CPU locking.

PiperOrigin-RevId: 481982546
 2022-10-18 11:59:29 -07:00
MediaPipe Team
f2821d840d Improve image classifier model maker documentation and replace the legacy code. 
PiperOrigin-RevId: 481979922
 2022-10-18 11:47:37 -07:00
MediaPipe Team
51879ae81a Remove unnecessary architectures from image classifier ModelSpec 
PiperOrigin-RevId: 481974529
 2022-10-18 11:29:11 -07:00
Yuqi Li
bc47589c9b Add model bundle in hand landmark task. 
PiperOrigin-RevId: 481960266
 2022-10-18 10:38:43 -07:00
Sebastian Schmidt
6bb5ff989d Fix #ifdef statement 
PiperOrigin-RevId: 481952827
 2022-10-18 10:13:09 -07:00
MediaPipe Team
47de64fc10 Open-sources TextToTensorCalculator. 
PiperOrigin-RevId: 481927522
 2022-10-18 08:28:53 -07:00
kinaryml
91b60da1dc Updated name for a test case 2022-10-18 05:35:05 -07:00
kinaryml
f166eb32e8 Updated names of test case methods 2022-10-18 04:34:17 -07:00
kinaryml
69f4daf9bc Removed unused BUILD 2022-10-18 04:31:35 -07:00
kinaryml
4932844410 Reverted changes to BUILD 2022-10-18 04:29:28 -07:00
kinaryml
f84e0bc1c6 Revised API implementation and added more tests for segment_for_video and segment_async 2022-10-18 04:24:12 -07:00
Kinar R
36ac0689d7
Merge branch 'google:master' into image-segmenter-python-impl 2022-10-18 15:13:44 +05:30
MediaPipe Team
ba93bab286 Fix ObjectDetector C++ flow limiter and improve documentation. 
PiperOrigin-RevId: 481774191
 2022-10-17 16:57:38 -07:00
Sebastian Schmidt
b09c0e9938 Internal change 
PiperOrigin-RevId: 481761007
 2022-10-17 15:56:55 -07:00
MediaPipe Team
58e5cc3c88 Changed the image to tensor converter interface such that the "Convert" 
function assumes the Tensor is preallocated before calling the function.

PiperOrigin-RevId: 481752678
 2022-10-17 15:20:25 -07:00
Camillo Lugaresi
a5e4219590 Internal change 
PiperOrigin-RevId: 481742662
 2022-10-17 14:39:17 -07:00
MediaPipe Team
cd32543786 Open-sources the bert_preprocessor_calculator_test. 
PiperOrigin-RevId: 481724320
 2022-10-17 13:30:11 -07:00
MediaPipe Team
5e543c506f Fix unused variable warnings 
Remove some unused variables and add unused attribute to other variables

PiperOrigin-RevId: 481706551
 2022-10-17 12:14:55 -07:00
MediaPipe Team
f6c2f86086 Fix -Wunused-variable warnings 
Remove unused variables. Add __unused attribute to variable used in assert clause. Move variable to inside header guard when only used in header guard.

PiperOrigin-RevId: 481699164
 2022-10-17 11:48:17 -07:00
Yuqi Li
660f1812c2 Internal change 
PiperOrigin-RevId: 481515490
 2022-10-16 17:23:01 -07:00
Jiuqiang Tang
2def7c8203 Add ".proto" suffix to gesture recognizer and object detector proto java package name. This aligns with the convention of image classifier and tasks core protos. 
PiperOrigin-RevId: 481451842
 2022-10-16 03:43:21 -07:00
MediaPipe Team
f7da95a130 Fix a typo in the description string 
PiperOrigin-RevId: 481421713
 2022-10-15 22:40:34 -07:00
MediaPipe Team
17202af6f7 Adds a preprocessor for Universal Sentence Encoder models. 
PiperOrigin-RevId: 481293992
 2022-10-14 21:38:28 -07:00
MediaPipe Team
5f3d5728e8 Open-source the regex_preprocessing_calculator 
PiperOrigin-RevId: 481256045
 2022-10-14 16:18:20 -07:00
Yuqi Li
eb52b72707 Internal change 
PiperOrigin-RevId: 481255129
 2022-10-14 16:13:13 -07:00
MediaPipe Team
ca28a19822 Open-source bert_preprocessor_calculator 
PiperOrigin-RevId: 481246966
 2022-10-14 15:31:31 -07:00
MediaPipe Team
42543f7ad6 Internal change 
PiperOrigin-RevId: 481227358
 2022-10-14 14:01:11 -07:00
Camillo Lugaresi
364b8e8106 Internal change 
PiperOrigin-RevId: 481210917
 2022-10-14 12:47:48 -07:00
MediaPipe Team
e283c10970 Internal change 
PiperOrigin-RevId: 481186706
 2022-10-14 11:04:43 -07:00
MediaPipe Team
0428550d75 Open source model_maker/python/core/tasks and model_maker/python/vision/image_classifier 
PiperOrigin-RevId: 481182271
 2022-10-14 10:47:34 -07:00
MediaPipe Team
6f3e8381ed Add support for rotation in ObjectDetector C++ API 
PiperOrigin-RevId: 481167472
 2022-10-14 09:47:54 -07:00
Kinar R
f160f28039
Merge branch 'google:master' into image-classification-python-impl 2022-10-14 15:31:01 +05:30
kinaryml
803210a86b Simplified async test cases to invoke the classifier in context 2022-10-14 03:00:29 -07:00
Jiuqiang Tang
0ebe6ccf59 Add filegroups to mediapipe java src dirs. 
PiperOrigin-RevId: 481080348
 2022-10-14 01:02:48 -07:00
Jiuqiang Tang
b632e645f5 Merge all BUILD files in the subdirectories of java/com/google/mediapipe/tasks/vision into one BUILD file. 
PiperOrigin-RevId: 481074268
 2022-10-14 00:15:40 -07:00
MediaPipe Team
9353ed6cce Java gesture recognizer Tasks API and unit test. 
PiperOrigin-RevId: 480978244
 2022-10-13 14:06:23 -07:00
MediaPipe Team
12c323ffde Exports gesture recognizer related proto as Java package. 
PiperOrigin-RevId: 480735444
 2022-10-12 15:41:31 -07:00
MediaPipe Team
179824a21d Use string_view when adding nodes/generators #cleanup 
PiperOrigin-RevId: 480721234
 2022-10-12 14:42:19 -07:00
MediaPipe Team
02746d0700 Remove unused includes. 
PiperOrigin-RevId: 480720274
 2022-10-12 14:38:04 -07:00
MediaPipe Team
a9fea36cb3 Put Destination::base_ into private section. (Cleanup.) 
PiperOrigin-RevId: 480678168
 2022-10-12 11:43:03 -07:00
MediaPipe Team
ae4b2ae577 Add support for input image rotation in ImageClassifier. 
PiperOrigin-RevId: 480676070
 2022-10-12 11:35:20 -07:00
MediaPipe Team
51a7606083 Add Java ImageClassifier API. 
PiperOrigin-RevId: 480656683
 2022-10-12 10:22:16 -07:00
Jiuqiang Tang
cbbd4718a0 Update mediapipe_aar.bzl to put more mediapipe framework java proto classes into AARs. 
PiperOrigin-RevId: 480583365
 2022-10-12 03:46:52 -07:00