* Remove assert

* debug0

* trace=not opt.sync

* sync to sync_bn fix

* Cleanup

* Close plots

* Replace fig.close() for plt.close()

Layer 21 includes the information of xsmall objects

* Auto-download Arial.ttf on init

* Fix ROOT

* Add Annotator() class

* Download Arial

* 2x for loop

* Cleanup

* tuple 2 list

* max_size=1920

* bold logging results to

* tolist()

* im = annotator.im

* PIL save in detect.py

* Smart asarray in detect.py

* revert to cv2.imwrite

* Cleanup

* Return result asarray

* Add `Profile()` profiler

* CamelCase Timeout

* Resize after mosaic

* pillow>=8.0.0

* daemon imwrite

* Add cv2 support

* Remove plot_wh_methods and plot_one_box

* pil=False for hubconf.py annotations

* im.shape bug fix

* colorstr common.py

* join daemons

* Update t.daemon

* Removed daemon saving

* Add `Profile()` profiler

* CamelCase Timeout

* Initial commit

* Update

* Add EarlyStopping feature

* Add comment

* Cleanup

* Cleanup2

* debug

* debug2

* debug3

* debug3

* debug4

* debug5

* debug6

* debug7

* debug8

* debug9

* debug10

* debug11

* debug12

* Cleanup

* Add TODO for known DDP issue

This PR brings alignment in AP computation practices with Detectron2 and MMDetection. 

Problem first noted by @yusiyoh in https://github.com/ultralytics/yolov5/issues/4546

Auto-fix corrupt JPEGs PR introduced a bug whereby the f.seek() operation read all of the bytes in the image, resulting in the PIL image having nothing to read upon the .save() operation. 

Fix was to re-open the image using PIL before saving.

* Autofix corrupt JPEGs

This PR automatically re-saves corrupt JPEGs and trains with the resaved images. WARNING: this will overwrite the existing corrupt JPEGs in a dataset and replace them with correct JPEGs, though the filesize may increase and the image contents may not be exactly the same due to lossy JPEG compression schemes. Results may vary by JPEG decoder and hardware.

Current behavior is to exclude corrupt JPEGs from training with a warning to the user, but many users have been complaining about large parts of their dataset being excluded from training.

* Clarify re-save reason

Profiling fix copies input to Detect layer to circumvent inplace changes to the feature maps.

* Auto TFLite uint8 detection

This PR automatically determines if TFLite models are uint8 quantized rather than accepting a manual argument.

The quantization determination is based on @zldrobit comment https://github.com/ultralytics/yolov5/pull/1127#issuecomment-901713847

* Cleanup

* Add `install=True` argument to `check_requirements`

* Update general.py

* Improve docstrings and run names

* default wandb login prompt with timeout

* return key

* Update api_key check logic

* Properly support zipped dataset feature

* update docstring

* Revert tuorial change

* extend changes to log_dataset

* add run name

* bug fix

* bug fix

* Update comment

* fix import check

* remove unused import

* Hardcore .yaml file extension

* reduce code

* Reformat using pycharm

* Remove redundant try catch

* More refactoring and bug fixes

* retry

* Reformat using pycharm

* respect LOGGERS include list

* Fix

* fix

* refactor constructor

* refactor

* refactor

* refactor

* PyCharm reformat
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

* `check_requirements(('coremltools',))`

* Update ci-testing.yml

* Update ci-testing.yml

* fix missing labels after augmentation

* Update datasets.py

Cleanup
Co-authored-by: Huu Quan <huuquan@HuuQuans-MacBook.local>
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

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

* Set auto=False for int8 calibration

* Update requirements.txt for TensorFlow and TFLite export

* Read anchors directly from PyTorch weights

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

* Remove check_anchor_order, check_file, set_logging from import

* Reformat code and optimize imports

* Autodownload model and check cfg

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

* Adjust representative_dataset

* Put representative dataset in tfl_int8 block

* detect.py TF inference

* weights to string

* weights to string

* cleanup tf.py

* Add --dynamic-batch-size

* Add xywh normalization to reduce calibration error

* Update requirements.txt

TensorFlow 2.3.1 -> 2.4.0 to avoid int8 quantization error

* Fix imports

Move C3 from models.experimental to models.common

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

* Set auto=False for int8 calibration

* Update requirements.txt for TensorFlow and TFLite export

* Read anchors directly from PyTorch weights

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

* Remove check_anchor_order, check_file, set_logging from import

* Reformat code and optimize imports

* Autodownload model and check cfg

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

* Adjust representative_dataset

* detect.py TF inference

* Put representative dataset in tfl_int8 block

* weights to string

* weights to string

* cleanup tf.py

* Add --dynamic-batch-size

* Add xywh normalization to reduce calibration error

* Update requirements.txt

TensorFlow 2.3.1 -> 2.4.0 to avoid int8 quantization error

* Fix imports

Move C3 from models.experimental to models.common

* implement C3() and SiLU()

* Fix reshape dim to support dynamic batching

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

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

* Add list support in check_img_size()

* Add list input support in detect.py

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

* Add int8 quantization support for TensorFlow 2.5

* Add get_coco128.sh

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

* Update requirements.txt

* Replace torch.load() with attempt_load()

* Update requirements.txt

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

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

* Cleanup after merge

* Cleanup2 after merge

* Cleanup3 after merge

* Add tf.py docstring with credit and usage

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

* Add use cases in docstring of tf.py

* Remove redundant `stride` definition

* Remove keras direct import

* Fix `check_requirements(('tensorflow>=2.4.1',))`
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

* Add `SPPF()` layer

* Cleanup

* Add credit

* Implement new headers

* Reformat 1

* Reformat 2

* Reformat 3 - math

* Reformat 4 - yaml

* Merge PIL and OpenCV box plotting functions

* Add ASCII check to plot_one_box

* Cleanup

* Cleanup2

* Remove `encoding='ascii'`

* Reinstate `encoding='ascii'` in emojis()

* Add yolov5s-ghost.yaml

* Finish C3Ghost

* Add C3Ghost to list

* Add C3Ghost to number of repeats if statement

* Fixes

* Cleanup