* :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>

* Updates

* Updates

* Updates

* Updates

* Updates

* Updates

* Updates

* Updates

* Updates

* Updates

* Cleanup

* Cleanup

Integer division by a float yields a (rounded) float. This causes
the dataloader to crash when creating a range.

* Move trainloader functions to class methods

* results = ThreadPool(NUM_THREADS).imap(self.load_image, range(n))

* Cleanup

* F541

* F821

* F841

* E741

* E302

* E722

* Apply suggestions from code review

* Update general.py

* Update datasets.py

* Update export.py

* Update plots.py

* Update plots.py

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

* Fix2 select_device() for Multi-GPU

* Cleanup

* Cleanup

* Simplify error message

* Improve assert

* Update torch_utils.py

* Add detect.py GIF video inference

* Cleanup

* Console corrupted -> corrupt 

Minor style changes.

* Update export.py

Video errors may occur.

When OpenCV retrieving image fail, original code would modify source images **inplace**, which may result in plotting bounding boxes on a black image. That is, before inference, source image `im0s[i]` is OK, but after inference before `Process predictions`,  `im0s[i]` may have been changed.

* WORLD_SIZE-safe workers

* Update with DDP comment

* `np.unique()` sort fix for segments

* Update datasets.py

* Fix float zeros format

* 255 to integer

* Update `set_logging()`

* Update export.py

* pre-commit fixes

* Update LoadImages

* Update LoadStreams

* Update print_args

* Single LOGGER definition

* yolo.py fix

Co-authored-by: pre-commit <pre-commit@example.com>

* define pre-commit

* add CI code

* configure

* apply pre-commit

* fstring

* apply MD

* pre-commit

* Update torch_utils.py

* Update print strings

* notes

* Cleanup code-format.yml

* Update setup.cfg

* Update .pre-commit-config.yaml

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

* take EXIF orientation tags into account when fixing corrupt images

* fit 120 char

* sort imports

* Update local exif_transpose comment

We have a local inplace version that is faster than the official as the image is not copied. AutoShape() uses this for Hub models, but here it is not important as the datasets.py usage is infrequent (AutoShape() it is applied every image).

* Update datasets.py

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

* take image files with uppercase extensions into account in autosplit

* case fix

* Refactor implementation

Removes additional variable (capital variable names are also only for global variables), and uses the same methodology as implemented earlier in datasets.py L409.

* Remove redundant rglob characters

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

* Autofix duplicate labels

PR changes duplicate label handling from report error and ignore image-label pair to report warning and autofix image-label pair. 

This should fix this common issue for users and allow everyone to get started and get a model trained faster and easier than before.

* sign fix

* Cleanup

* Increment cache version

* all to any fix

* Add train class filter feature to datasets.py

Allows for training on a subset of total classes if `include_class` list is defined on datasets.py L448:
```python
        include_class = []  # filter labels to include only these classes (optional)
```

* segments fix

* Re-open IP-camera videostream if disconnected. Provide IP-streaming analysis stability

* Re-open IP-camera videostream if disconnected. Provide IP-streaming analysis stability

* `self.imgs[i] *= 0` bug fix and cleanup

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

* Replace `os.system('unzip file.zip')` -> `ZipFile.extractall()`

* Cleanup

PR should produce datasets sorted alphabetically by filename. Cache version incremented to 0.5. 

Note: will force a one-time re-caching of existing datasets on first-use.