Duplicate of #3495 merged to `develop`. This PR will be merged to `master`.

Fixes https://github.com/ultralytics/yolov5/issues/3494.

Fix for https://github.com/ultralytics/yolov5/issues/958#issuecomment-849512083

* Add URL file download to check_file()

* cleanup

* pathlib bug fix

* Update cache v0.2 to include parent hash

Possible fix for https://github.com/ultralytics/yolov5/issues/3349

* Update datasets.py

* `@torch.no_grad()` decorator

* Update detect.py

Add try: except clause on missing requirements install to catch install failures. Prompted by Ultralytics YOLOv5 API failure on Cython.

Ultralytics REST API fix.

* Changing save_one_box 

Made to work with other changes to common.py

* PEP8 and single line BGR
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

* adds latest tag to match wandb defaults

* adds entity handling, 'last' tag

* fixes bug causing finished runs to resume

* removes redundant "last" tag for wandb artifact

* YouTube stream ending fix

Properly terminates YouTube streams on video end. Should resolve issues #2769 and #3220.

* Update datasets.py

Bug fix to check batch_size divisibility of utilized CUDA device count vs total system CUDA device count.

* Check CoreML models.train() mode

* Update export.py

Cython required by pycocotools in certain cases, i.e. https://github.com/cocodataset/cocoapi/issues/172

* Color can be none by default

* `plot_one_box()` default `color=(128, 128, 128)`
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

* Added max_det parameters in various places

* 120 character line

* PEP8

* 120 character line

* Update inference default to 1000 instances

* Update inference default to 1000 instances
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

* Allow to manual selection of device for torchhub models

* single line device

nested torch.device(torch.device(device)) ok
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

* rename class autoShape -> AutoShape

follow other class naming convention

* rename class autoShape -> AutoShape

follow other classes' naming convention

* rename class autoShape -> AutoShape

* SKU-110K CVPR2019 Dataset Auto-Download

This PR adds the **SKU-110K CVPR2019 Dataset** https://github.com/eg4000/SKU110K_CVPR19 to the list of YOLOv5 🚀  auto-download list.

- [x] [COCO](https://cocodataset.org/#home)
- [x] [COCO128](https://www.kaggle.com/ultralytics/coco128)
- [x] [VOC](http://host.robots.ox.ac.uk/pascal/VOC/)
- [x] [Argoverse-HD](http://www.cs.cmu.edu/~mengtial/proj/streaming/)
- [x] [VisDrone2019-DET](https://github.com/VisDrone/VisDrone-Dataset)
- [x] [GlobalWheat-2020](http://www.global-wheat.com/)
- [ ] [KITTI](https://www.kaggle.com/twaldo/kitti-object-detection)
- [x] [SKU110](https://github.com/eg4000/SKU110K_CVPR19) (grocery store items)
- [ ] [CityScapes](https://www.cityscapes-dataset.com/)
- [x] [Objects365](https://www.objects365.org/overview.html)
- [ ] [OpenImages](https://storage.googleapis.com/openimages/web/index.html)

```yaml
# SKU-110K retail items dataset https://github.com/eg4000/SKU110K_CVPR19
# Train command: python train.py --data SKU-110K.yaml
# Default dataset location is next to YOLOv5:
#   /parent_folder
#     /datasets/SKU-110K
#     /yolov5
```

* Update SKU-110K.yaml

* Update SKU-110K.yaml

* Scope all hubconf.py imports for torch.hub.list()

* Update hubconf.py

* Allow users to skip exporting in formats that they don't care about

* Correct comments

* Update export.py

renamed --skip-format to --exclude

* Switched format from exclude to include (as instructed by @glenn-jocher)

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