fix(utils,exp): logger compat issue and exp check (#1618)

feat(exp): sequence support for exp options merge function

feat(tools): add assignment visualizer (#1616)

fix(data): read img bug in voc dataset

feat(data): support custom dataset cache

chore(workflow): remove py36 check

fix(models): no-candidate anchor issue for tiny objects during label assign

chore(tools): upgrade onnxsim to v0.4.1

feat(data): support cache ram of COCO dataset

doc(ncnn): update README

style(models): fix style error

Avoid ScatterND ops that will cause error in openvino model optimizer.

feat(demo): add nebullvm inference accelerator

feat(logger): W&B logger with VOC datasets

fix(README): invalid link of nano model

chore(tools): add decode_in_inference option to export_torchscript

feat(hub): loading a custom model with `torch.hub.load` 

chore(logger): log predictions during training to wandb tables 

in memory of Dr. Jian Sun

fix(demo): typo in README

fix(wandb): key messed in wandb option parsing

docs(README): use absolute URLs for model names

chore(README): add StreamYOLO.

tag(YOLOX): bump to 0.3

fix(README & model) some typos

fix(README):  typo

fix(demo/TensorRT): except in logging.h

fix(README): link to `yolox_base.py`

fix(README):  correct onnx url path of yolox_x

feat(model): support hub load