Evaluation of 'best' and 'last' models will use the same params as the evaluation during the training phase. 
This PR fixes https://github.com/ultralytics/yolov5/issues/3907

* Update `dataset_stats()` for zipped datasets

@KalenMike

* cleanup

Fix inference from PIL source.

* Feature visualization update

* Save to jpg (faster)

* Save to png

* Create hubconf.py

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

* Albumentations integration

* ToGray p=0.01

* print confirmation

* create instance in dataloader init method

* improved version handling

* transform not defined fix

* assert string update

* create check_version()

* add spaces

* update class comment

* Hub models `map_location=device`

* cleanup

Fix for `KeyError: 'anchors'` error when start hyperparameter evolution:
```bash
python train.py --evolve
```

```bash
Traceback (most recent call last):
  File "E:\yolov5\train.py", line 623, in <module>
    hyp[k] = max(hyp[k], v[1])  # lower limit
KeyError: 'anchors'
```

* Update BGR2RGB ops

* speed improvements

* cleanup

* Create `utils/augmentations.py`

* cleanup

* Added the recording feature for multiple streams

Thanks for the very cool repo!!
I was trying to record multiple feeds at the same time, but the current version of the detector only had one video writer and one vid_path!
So the streams were not being saved and only were initialized with one frame and this process didn't record the whole thing.

Fix:
I made a list of `vid_writer` and `vid_path` and the `i` from the loop over the `pred` took care of the writer which need to work!

I hope this helps, Thanks!

* Cleanup list lengths

* batch size variable

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

* evolve command accepts argument for number of generations

* evolve generations argument used in evolve for loop

* evolve argument boolean fixes

* default to 300 evolve generations

* Update train.py
Co-authored-by: John San Soucie <jsansoucie@whoi.edu>
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

* rotating an image according to its exif tag

* Update common.py

* Update datasets.py

* Update datasets.py

faster

* delete extraneous gpg file

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

* Copy-paste augmentation initial commit

* if any segments

* Add obscuration rejection

* Add copy_paste hyperparameter

* Update comments

* Update datasets.py to read every 4th frame of streams

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

@KalenMike

* Update `feature_visualization()`

Only plot for data with height, width > 1

* cleanup

* Cleanup

* Add feature map visualization

Add a feature_visualization function to visualize the mid feature map of the model.

* Update yolo.py

* remove boolean from forward and reorder if statement

* remove print from forward

* General cleanup

* Indent

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

* gradient accumulation during warmup in train.py

Context:
`accumulate` is the number of batches/gradients accumulated before calling the next optimizer.step().
During warmup, it is ramped up from 1 to the final value nbs / batch_size. 
Although I have not seen this in other libraries, I like the idea. During warmup, as grads are large, too large steps are more of on issue than gradient noise due to small steps.

The bug:
The condition to perform the opt step is wrong
> if ni % accumulate == 0:
This produces irregular step sizes if `accumulate` is not constant. It becomes relevant when batch_size is small and `accumulate` changes many times during warmup.

This demo also shows the proposed solution, to use a ">=" condition instead:
https://colab.research.google.com/drive/1MA2z2eCXYB_BC5UZqgXueqL_y1Tz_XVq?usp=sharing

Further, I propose not to restrict the number of warmup iterations to >= 1000. If the user changes hyp['warmup_epochs'], this causes unexpected behavior. Also, it makes evolution unstable if this parameter was to be optimized.

* replace last_opt_step tracking by do_step(ni)

* add docstrings

* move down nw

* Update train.py

* revert math import move
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

* link to the competition repaired

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

For seed=0 on single-GPU.

* merge master

* Update get_coco.sh