Code

Add TensorFlow and TFLite export#1127

Code

Merged

glenn-jocher merged 71 commits intoultralytics:masterultralytics/yolov5:masterfrom zldrobit:tf-only-exportzldrobit/yolov5:tf-only-exportCopy head branch name to clipboardAug 17, 2021

+626-17

Commits

Commits on Nov 28, 2020

• Add models/tf.py for TensorFlow and TFLite export

  

  zldrobit

  authored and

  glenn-jocher

  committed
  5b6528c

  Copy full SHA for 5b6528c

• Set auto=False for int8 calibration

  

  zldrobit

  authored and

  glenn-jocher

  committed
  a30dad4

  Copy full SHA for a30dad4

• Update requirements.txt for TensorFlow and TFLite export

  

  zldrobit

  authored and

  glenn-jocher

  committed
  f0cb6e2

  Copy full SHA for f0cb6e2

• Read anchors directly from PyTorch weights

  

  zldrobit

  authored and

  glenn-jocher

  committed
  ce73d3d

  Copy full SHA for ce73d3d

• Add --tf-nms to append NMS in TensorFlow SavedModel and GraphDef export

  

  zldrobit

  authored and

  glenn-jocher

  committed
  d101f7e

  Copy full SHA for d101f7e

• Remove check_anchor_order, check_file, set_logging from import

  

  zldrobit

  authored and

  glenn-jocher

  committed
  55fb2eb

  Copy full SHA for 55fb2eb

• Reformat code and optimize imports

  glenn-jocher

  committed
  3dd69d9

  Copy full SHA for 3dd69d9

• Autodownload model and check cfg

  glenn-jocher

  committed
  e8a9ad2

  Copy full SHA for e8a9ad2

• update --source path, img-size to 320, single output

  glenn-jocher

  committed
  efbb853

  Copy full SHA for efbb853

• Adjust representative_dataset

  glenn-jocher

  committed
  aed53ce

  Copy full SHA for aed53ce

Commits on Nov 30, 2020

• Put representative dataset in tfl_int8 block

  zldrobit

  committed
  ccb2336

  Copy full SHA for ccb2336

Commits on Dec 2, 2020

• detect.py TF inference

  glenn-jocher

  committed
  9f893c8

  Copy full SHA for 9f893c8

• Merge remote-tracking branch 'origin/tf-only-export' into tf-only-export

  glenn-jocher

  committed
  d9fad06

  Copy full SHA for d9fad06

• weights to string

  glenn-jocher

  committed
  49a9e05

  Copy full SHA for 49a9e05

• weights to string

  glenn-jocher

  committed
  1867bb4

  Copy full SHA for 1867bb4

Commits on Dec 4, 2020

• cleanup tf.py

  glenn-jocher

  committed
  4eed608

  Copy full SHA for 4eed608

Commits on Dec 22, 2020

• Add --dynamic-batch-size

  zldrobit

  committed
  8ba2ca9

  Copy full SHA for 8ba2ca9

• Add xywh normalization to reduce calibration error

  zldrobit

  committed
  4d9104a

  Copy full SHA for 4d9104a

• Merge branch 'master' into tf-only-export

  glenn-jocher

  authored
  ae9bce8

  Copy full SHA for ae9bce8

Commits on Dec 23, 2020

• Update requirements.txt

  Show description for cabb802
  zldrobit

  committed
  cabb802

  Copy full SHA for cabb802

Commits on Dec 24, 2020

• Fix imports

  Show description for a5967f8
  zldrobit

  committed
  a5967f8

  Copy full SHA for a5967f8

Commits on Jan 8, 2021

• Add models/tf.py for TensorFlow and TFLite export

  

  zldrobit

  authored and

  glenn-jocher

  committed
  dbc7f71

  Copy full SHA for dbc7f71

• Set auto=False for int8 calibration

  

  zldrobit

  authored and

  glenn-jocher

  committed
  565e620

  Copy full SHA for 565e620

• Update requirements.txt for TensorFlow and TFLite export

  

  zldrobit

  authored and

  glenn-jocher

  committed
  fc26561

  Copy full SHA for fc26561

• Read anchors directly from PyTorch weights

  

  zldrobit

  authored and

  glenn-jocher

  committed
  05cc389

  Copy full SHA for 05cc389

• Add --tf-nms to append NMS in TensorFlow SavedModel and GraphDef export

  

  zldrobit

  authored and

  glenn-jocher

  committed
  817fcf8

  Copy full SHA for 817fcf8

• Remove check_anchor_order, check_file, set_logging from import

  

  zldrobit

  authored and

  glenn-jocher

  committed
  9a4ce5e

  Copy full SHA for 9a4ce5e

• Reformat code and optimize imports

  glenn-jocher

  committed
  719479e

  Copy full SHA for 719479e

• Autodownload model and check cfg

  glenn-jocher

  committed
  243fa7f

  Copy full SHA for 243fa7f

• update --source path, img-size to 320, single output

  glenn-jocher

  committed
  3b6cf12

  Copy full SHA for 3b6cf12

• Adjust representative_dataset

  glenn-jocher

  committed
  5a5f949

  Copy full SHA for 5a5f949

• detect.py TF inference

  glenn-jocher

  committed
  00d50fd

  Copy full SHA for 00d50fd

• Put representative dataset in tfl_int8 block

  

  zldrobit

  authored and

  glenn-jocher

  committed
  061907b

  Copy full SHA for 061907b

• weights to string

  glenn-jocher

  committed
  ca4550b

  Copy full SHA for ca4550b

• weights to string

  glenn-jocher

  committed
  5e04c5c

  Copy full SHA for 5e04c5c

• cleanup tf.py

  glenn-jocher

  committed
  9121a87

  Copy full SHA for 9121a87

• Add --dynamic-batch-size

  

  zldrobit

  authored and

  glenn-jocher

  committed
  e9bc606

  Copy full SHA for e9bc606

• Add xywh normalization to reduce calibration error

  

  zldrobit

  authored and

  glenn-jocher

  committed
  dacd8af

  Copy full SHA for dacd8af

• Update requirements.txt

  Show description for b492af9
  

  zldrobit

  authored and

  glenn-jocher

  committed
  b492af9

  Copy full SHA for b492af9

• Fix imports

  Show description for fbf5a45
  

  zldrobit

  authored and

  glenn-jocher

  committed
  fbf5a45

  Copy full SHA for fbf5a45

• Merge branch 'tf-only-export' of https://github.com/zldrobit/yolov5 into tf-only-export

  glenn-jocher

  committed
  c4cfbd9

  Copy full SHA for c4cfbd9

• implement C3() and SiLU()

  glenn-jocher

  committed
  c761637

  Copy full SHA for c761637

Commits on Feb 2, 2021

• Fix reshape dim to support dynamic batching

  zldrobit

  committed
  36ed3cd

  Copy full SHA for 36ed3cd

Commits on Feb 4, 2021

• Merge branch 'master' into tf-only-export

  glenn-jocher

  authored
  aad9e24

  Copy full SHA for aad9e24

Commits on Feb 6, 2021

• Merge branch 'master' into tf-only-export

  glenn-jocher

  authored
  8ec975e

  Copy full SHA for 8ec975e

Commits on Mar 12, 2021

• Add epsilon argument in tf_BN, which is different between TF and PT

  zldrobit

  committed
  b0fa5a3

  Copy full SHA for b0fa5a3

• Set stride to None if not using PyTorch, and do not warmup without PyTorch

  zldrobit

  committed
  710bf56

  Copy full SHA for 710bf56

Commits on Mar 23, 2021

• Add list support in check_img_size()

  zldrobit

  committed
  c45ceef

  Copy full SHA for c45ceef

• Add list input support in detect.py

  zldrobit

  committed
  0d39b24

  Copy full SHA for 0d39b24

Commits on Mar 25, 2021

• merge ultralytics:master

  glenn-jocher

  committed
  47da942

  Copy full SHA for 47da942

Commits on Mar 28, 2021

• merge ultralytics:master

  glenn-jocher

  committed
  8cb7032

  Copy full SHA for 8cb7032

• sys.path.append('./') to run from yolov5/

  glenn-jocher

  committed
  4e1485b

  Copy full SHA for 4e1485b

Commits on Apr 3, 2021

• Add int8 quantization support for TensorFlow 2.5

  zldrobit

  committed
  e4e6d6f

  Copy full SHA for e4e6d6f

Commits on May 6, 2021

• Add get_coco128.sh

  zldrobit

  committed
  aafe224

  Copy full SHA for aafe224

• Remove --no-tfl-detect in models/tf.py (Use tf-android-tfl-detect branch for EdgeTPU)

  zldrobit

  committed
  d3be281

  Copy full SHA for d3be281

Commits on Jun 2, 2021

• Merge branch 'develop' into tf-only-export

  glenn-jocher

  authored
  9ca5d7a

  Copy full SHA for 9ca5d7a

• Update requirements.txt

  glenn-jocher

  authored
  215c865

  Copy full SHA for 215c865

Commits on Jun 7, 2021

• Replace torch.load() with attempt_load()

  zldrobit

  committed
  a2867da

  Copy full SHA for a2867da

• Update requirements.txt

  zldrobit

  committed
  86768d1

  Copy full SHA for 86768d1

Commits on Jun 11, 2021

• Add --tf-raw-resize to set half_pixel_centers=False

  zldrobit

  committed
  972c6a2

  Copy full SHA for 972c6a2

Commits on Jun 18, 2021

• Add --agnostic-nms for TF class-agnostic NMS

  zldrobit

  committed
  eed4980

  Copy full SHA for eed4980

Commits on Aug 16, 2021

• Merge master

  glenn-jocher

  committed
  34b7d67

  Copy full SHA for 34b7d67

• Cleanup after merge

  glenn-jocher

  committed
  10eebf2

  Copy full SHA for 10eebf2

• Cleanup2 after merge

  glenn-jocher

  committed
  e3aa755

  Copy full SHA for e3aa755

• Cleanup3 after merge

  glenn-jocher

  committed
  27fc39b

  Copy full SHA for 27fc39b

• Add tf.py docstring with credit and usage

  glenn-jocher

  committed
  48a6bf8

  Copy full SHA for 48a6bf8

Commits on Aug 17, 2021

• pb saved_model and tflite use only one model in detect.py

  zldrobit

  committed
  4e12c70

  Copy full SHA for 4e12c70

• Add use cases in docstring of tf.py

  zldrobit

  committed
  ea0274f

  Copy full SHA for ea0274f

• Remove redundant stride definition

  glenn-jocher

  committed
  2f63c6d

  Copy full SHA for 2f63c6d

• Remove keras direct import

  glenn-jocher

  committed
  c3f46c3

  Copy full SHA for c3f46c3

• Fix check_requirements(('tensorflow>=2.4.1',))

  glenn-jocher

  committed
  3133381

  Copy full SHA for 3133381