* train.py GPU memory fix

* ema

* cuda

* cuda

* zeros input

* to device

* batch index 0

@kalenmike this is a PR to add image filenames and labels to our stats dictionary and to save the dictionary to JSON. Save location is next to the train labels.cache file. The single JSON contains all stats for entire dataset.

Usage example:
```python
from utils.datasets import *

dataset_stats('coco128.yaml', verbose=True)
```

* Update README.md

* added hosted images

* added new logo

* testing image hosting

* changed svgs to pngs

* removed old header

* Update README.md

* correct colab image source

* splash.jpg

* rocket and W&B fix

* added contributing template

* added social media to top section

* increased size of top social media

* cleanup and updates

* rearrange quickstarts

* API cleanup

* PyTorch Hub cleanup

* Add tutorials

* cleanup

* update CONTRIBUTING.md

* Update README.md

* update wandb link

* Update README.md

* remove tutorials header

* update environments and integrations

* Comment API image

* Update README.md

* Update README.md

* Update README.md

* Update README.md

* Update README.md

* double spaces after section

* Update README.md

* Update README.md

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

* 'changes-in_dataset'

* Update datasets.py

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

* supress wandb images size mismatch warning

* supress wandb images size mismatch warning

* PEP8 reformat and optimize imports

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

* Update `check_file()`

* Update datasets.py

* premature end of JPEG images

* PEP8 reformat

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

* Update CONTRIBUTING.md

* Update CONTRIBUTING.md

* Update CONTRIBUTING.md

* Update CONTRIBUTING.md

Be more user-friendly to new users

* Update `dataset_stats()` to list of dicts

@kalenmike

* Update datasets.py

* Remove `/weights` directory

* cleanup

…3668)

* Cache v0.3: improved corrupt image/label reporting

Fix for ultralytics#3656 (comment)

* cleanup

* EMA changes for pre-model's batch_size

* Update train.py

* Update torch_utils.py

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

Swapped order of operations for faster first per ultralytics@f527704#r52362419

…#3686)

* Create shortcircuit in augment_hsv when hyperparameter are zero

* implement faster opt-in

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

* Slightly modify CLI execution

This simple change makes it easier to run the primary functions of this
repo (train/detect/test) from within Python. An object which represents
`opt` can be constructed and fed to the `main` function of each of these
modules, rather than having to call the lower level functions directly,
or run the module as a script.

* Update export.py

Add CLI parsing update for more convenient module usage within Python.

Co-authored-by: Lewis Belcher <lb@desupervised.io>

…cs#3680)

* Update DDP for `torch.distributed.run`

* Add LOCAL_RANK

* remove opt.local_rank

* backend="gloo|nccl"

* print

* print

* debug

* debug

* os.getenv

* gloo

* gloo

* gloo

* cleanup

* fix getenv

* cleanup

* cleanup destroy

* try nccl

* return opt

* add --local_rank

* add timeout

* add init_method

* gloo

* move destroy

* move destroy

* move print(opt) under if RANK

* destroy only RANK 0

* move destroy inside train()

* restore destroy outside train()

* update print(opt)

* cleanup

* nccl

* gloo with 60 second timeout

* update namespace printing

* Eliminate `total_batch_size` variable

* cleanup

* Update train.py

* Update train.py explicit arguments

* Update train.py

* Add run method

* Allow config cahnge

* Allow val change in wandb config

* Don't resume transfer learning runs

* Add entity in log dataset

* Add yolov5/ to path

* rename functions to run()

* cleanup

* rename fix

* CI fix

* cleanup find models/export.py

* Fix `img2label_paths()` order

* fix, 1

@kalenmike

Fix for ultralytics#3741

edit comment

* Add optional dataset.yaml `path` attribute

@kalenmike

* pass locals to python scripts

* handle lists

* update coco128.yaml

* Capitalize first letter

* add test key

* finalize GlobalWheat2020.yaml

* finalize objects365.yaml

* finalize SKU-110K.yaml

* finalize SKU-110K.yaml

* finalize VisDrone.yaml

* NoneType fix

* update download comment

* voc to VOC

* update

* update VOC.yaml

* update VOC.yaml

* remove dashes

* delete get_voc.sh

* force coco and coco128 to ../datasets

* Capitalize Argoverse_HD.yaml

* Capitalize Objects365.yaml

* update Argoverse_HD.yaml

* coco segments fix

* VOC single-thread

* update Argoverse_HD.yaml

* update data_dict in test handling

* create root

* Edit Comments for numpy2torch tensor process

Edit Comments for numpy2torch tensor process

* add xyxy2xywhn

add xyxy2xywhn

* add xyxy2xywhn

* formatting

* pass arguments

pass arguments

* edit comment for xyxy2xywhn()

edit comment for xyxy2xywhn()

* cleanup datasets.py

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

…#3772)

* Update tables API and windows path fix

* update dataset check

* update clip_coords for numpy

* uncomment

* cleanup

* Add autosplits

* fix

* cleanup

Co-authored-by: U-LAPTOP-5N89P8V7\banhu <ban.huang@foxmail.com>

* merge master

* Update get_coco.sh

For seed=0 on single-GPU.

* link to the competition repaired

* Update README.md

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>

* 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>

* Update `feature_visualization()`

Only plot for data with height, width > 1

* cleanup

* Cleanup

@kalenmike

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

* Update datasets.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

* 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>

* 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>

* 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>

* Create `utils/augmentations.py`

* cleanup

* Update BGR2RGB ops

* speed improvements

* 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'
```

* Hub models `map_location=device`

* cleanup

* 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

* Create hubconf.py

* Add save_dir variable

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

* Feature visualization update

* Save to jpg (faster)

* Save to png