* :tada: :new: now can read Chinese image path. 

use "cv2.imdecode(np.fromfile(f, np.uint8), cv2.IMREAD_COLOR)" instead of "cv2.imread(f)" for Chinese image path.

* Update datasets.py

* Update __init__.py

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

* Multi-threaded VOC download

* Update VOC.yaml

* Update

* Update general.py

* Update general.py

* Remove named arguments where possible

Speed improvements.

* Update yolo.py

* Update yolo.py

* Update yolo.py

Stems not working correctly for YOLOv5l with current .rstrip() implementation. After fix:
```
YOLOv5l summary: 468 layers, 46563709 parameters, 46563709 gradients, 109.3 GFLOPs
```

* Update loss.py with `if self.gr < 1:`

* Update loss.py

Improved profiling.

* Update `check_anchor_order()`

Use mean area per output layer for added stability.

* Check in pixel-space not grid-space fix

* use pillow to save higher-quality jpg (w/o color subsampling)

* Cleanup and doc issue

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

Resolves git command errors that currently happen in image, i.e.:

```bash
root@382ae64aeca2:/usr/src/app# git pull
Warning: Permanently added the ECDSA host key for IP address '140.82.113.3' to the list of known hosts.
git@github.com: Permission denied (publickey).
fatal: Could not read from remote repository.

Please make sure you have the correct access rights
and the repository exists.
```

* device as class attribute

* Update loss.py

* Update loss.py

* improve zeros

* tensor split

* fix: add default font as fallback

Add default font as fallback if the downloading of the Arial.ttf font
fails for some reason, e.g. no access to public internet.

* Update plots.py

Co-authored-by: Maximilian Strobel <Maximilian.Strobel@infineon.com>
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

* Conditional `Timeout()` by OS (disable on Windows)

* Update general.py

* Add benchmarks.py GPU support

* Updates

* Updates

* Updates

* Updates

* Add --half

* Add TRT requirements

* Cleanup

* Add TF to warmup types

* Update export.py

* Update export.py

* Update benchmarks.py

* removed transpose op for better edgetpu support

* fix for training case

* enabled experimental new quantizer flag

* precalculate add and mul ops at compile time

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

* Update common.py speed improvements

Eliminate .to() ops where possible for reduced data transfer overhead. Primarily affects warmup and PyTorch Hub inference.

* Updates

* Updates

* Update detect.py

* Update val.py

Fixes ONNX validation that returns outputs on CPU.

* New val.py `cuda` variable

Fix for ONNX GPU val.

* Update val.py

* DetectMultiBackend() `--half` handling

* CI fixes

* rename .half to .fp16 to avoid conflict

* warmup fix

* val update

* engine update

* engine update

* Move `git_describe()` to general.py

* Move `git_describe()` to general.py

* Fix TRT `max_workspace_size` deprecation notice

* Update export.py

* Update export.py

* Update yolov5s.yaml

* Update yolov5s.yaml

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

* Updates

* Updates

* Updates

* Updates

* Updates

* Updates

* Updates

* Updates

* Updates

* Updates

* Cleanup

* Cleanup

* Update sweep.yaml

Changed focal loss gamma search range between 1 and 4

* Update sweep.yaml

lowered the min value to match default