* Youtube livestream detection

* dependancy update to auto install pafy

* Remove print

* include youtube_dl in deps

* PEP8 reformat

* youtube url check fix

* reduce lines

* add comment

* update check_requirements

* stream framerate fix

* Update README.md

* cleanup

* PEP8

* remove cap.retrieve() failure code
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

This fix should allow for visualizing YOLOv5 model graphs correctly in Tensorboard by uncommenting line 335 in train.py:
```python
                    if tb_writer:
                        tb_writer.add_graph(torch.jit.trace(model, imgs, strict=False), [])  # add model graph
```

The problem was that the detect() layer checks the input size to adapt the grid if required, and tracing does not seem to like this shape check (even if the shape is fine and no grid recomputation is required). The following will warn:
https://github.com/ultralytics/yolov5/blob/0cae7576a9241110157cd154fc2237e703c2719e/train.py#L335

Solution is below. This is a YOLOv5s model displayed in TensorBoard. You can see the Detect() layer merging the 3 layers into a single output for example, and everything appears to work and visualize correctly.
```python
tb_writer.add_graph(torch.jit.trace(model, imgs, strict=False), [])
```
<img width="893" alt="Screenshot 2021-04-11 at 01 10 09" src="https://user-images.githubusercontent.com/26833433/114286928-349bd600-9a63-11eb-941f-7139ee6cd602.png">

* wandb_logging PEP8 reformat

* Update wandb_utils.py

PR https://github.com/ultralytics/yolov5/pull/2725 introduced a very specific bug that only affects multi-GPU trainings. Apparently the cause was using the torch.cuda.amp decorator in the autoShape forward method. I've implemented amp more traditionally in this PR, and the bug is resolved.

* Add Hub results.pandas() method

New method converts results from torch tensors to pandas DataFrames with column names.

This PR may partially resolve issue https://github.com/ultralytics/yolov5/issues/2703

```python
results = model(imgs)

print(results.pandas().xyxy[0])
         xmin        ymin        xmax        ymax  confidence  class    name
0   57.068970  391.770599  241.383545  905.797852    0.868964      0  person
1  667.661255  399.303589  810.000000  881.396667    0.851888      0  person
2  222.878387  414.774231  343.804474  857.825073    0.838376      0  person
3    4.205386  234.447678  803.739136  750.023376    0.658006      5     bus
4    0.000000  550.596008   76.681190  878.669922    0.450596      0  person
```

* Update comments 

torch example input now shown resized to size=640 and also now a multiple of P6 stride 64 (see https://github.com/ultralytics/yolov5/issues/2722#issuecomment-814785930)

* apply decorators

* PEP8

* Update common.py

* pd.options.display.max_columns = 10

* Update common.py

* Update README with collapsable notes.

* cleanup

* center table

Fix 2 for 'Model predict with forward will fail if PIL image does not have filename attribute' #2702

Proposed fix for 'Model predict with forward will fail if PIL image does not have filename attribute' #2702

Requested in https://github.com/ultralytics/yolov5/issues/2686

Slight speedup.

* PyTorch Hub model.save() increment as runs/hub/exp

This chane will align PyTorch Hub results saving with the existing unified results saving directory structure of

runs/
  /train
  /detect
  /test
  /hub
    /exp
    /exp2
    ...

* cleanup

* Set resume flag to false

* Check existance of val dataset

* yolotr

* transformer block

* Remove bias in Transformer

* Remove C3T

* Remove a deprecated class

* put the 2nd LayerNorm into the 2nd residual block

* move example model to models/hub, rename to -transformer

* Add module comments and TODOs

* Remove LN in Transformer

* Add comments for Transformer

* Solve the problem of MA with DDP

* cleanup

* cleanup find_unused_parameters

* PEP8 reformat
Co-authored-by: DingYiwei <846414640@qq.com>
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

* Update README with Tips for Best Results tutorial

* Update README.md

* Add label smoothing option

* Correct data type

* add_log

* Remove log

* Add log

* Update loss.py

remove comment (too versbose)
Co-authored-by: phattran <phat.tranhoang@cyberlogitec.com>
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

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

* Add opencv-contrib-python to requirements.txt

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

* Speed profiling improvements

* Update torch_utils.py

deepcopy() required to avoid adding elements to model.

* Update torch_utils.py

When using tqdm, sometimes it can't print in one line and roll to next line.

I think this should help speed up CUDA inference, as currently models may be running in FP32 inference mode on CUDA devices unnecesarily.

Fix for #2630 raised by @Pro100rus32

Add stderr=subprocess.STDOUT to catch error messages.

* W&B resume ddp from run link fix

* Native DDP W&B support for training, resuming

Catch 'fatal: not a git repository' returns and return '' instead (observed in GCP Hub checks).

Update Docker FROM nvcr.io/nvidia/pytorch:21.03-py3

Attempt to resolve tensorboard Docker error in https://github.com/ultralytics/yolov5/issues/2573

```python
        self.n = len(self.pred)  # number of images (batch size)
```

* Fix Indentation in test.py

* CI fix

* Comply with PEP8: 80 characters per line

* Update detections() self.t = tuple()

Fix multiple results.print() bug.

* Update experimental.py

* Update yolo.py

Updated device selection string with fallback for non-git directories.
```python
def select_device(device='', batch_size=None):
    # device = 'cpu' or '0' or '0,1,2,3'
    s = f'YOLOv5 🚀 {git_describe() or date_modified()} torch {torch.__version__} '  # string
...
```