Support upstream ultralytics v4.0 release

This is a minor release to support the upstream ultralytics/yolov5 v4.0 release stacks

• Add export friendly substitutions of SiLU (#69)
• Support ultralytics released v4.0 stacks (#66)

TVM and Lightning inference support

Highlights

Support TVM backend inference

This release support TVM backend inference (#63, #53, #51, #50). Here is a rough comparison of inference time consumed on Jupyter notebook (iPython) (#54).

• On the ONNXRuntime backend,

  CPU times: user 2.04 s, sys: 0 ns, total: 2.04 s
  Wall time: 55.8 ms

• On the TorchScript backend,

  CPU times: user 2.03 s, sys: 32 ms, total: 2.06 s
  Wall time: 60.5 ms

• On the PyTorch backend,

  CPU times: user 3.87 s, sys: 60 ms, total: 3.93 s
  Wall time: 116 ms

• On the TVM backend,

  CPU times: user 528 ms, sys: 364 ms, total: 892 ms
  Wall time: 22.3 ms

Inherited from Lightning

yolort module are now inherited from LightningModule (#62, #48, #46, #43). To read a source of image(s) and detect its objects just run as following.

from yolort.models import yolov5s

# Load model
model = yolov5s(pretrained=True, score_thresh=0.45)
model.eval()

# Perform inference on an image file
predictions = model.predict('bus.jpg')
# Perform inference on a list of image files
predictions = model.predict(['bus.jpg', 'zidane.jpg'])

Backwards Incompatible Changes

• Make yolort as a Python package (#55, #56)

New Features

• Add flake8 lint unittest (#58)
• Create codeql-analysis.yml (#37)
• Refactor darknet backbone as separate modules (#33)
• Add ISSUE_TEMPLATE and codecov workflows (#24)
• Refactor model unittest (#23)
• Replacing all torch.jit.annotations with typing (#22)
• Add automatic rebase (#11)
• Add torchscript export and model unittest (#8)
• Support yolov5m and yolov5l models (#7)
• Add essential scripts for training (#5)
• Enable unittest in pytorch nightly version (#4)
• Add half precision inference in libtorch (#1)

Bug Fixes

• Rescale to original scale after post-processor (#47)
• Suppress cpp language analysis (#40)
• Fix inference inconsistency compare with ultralytics (#31)
• Fix loss computation (#25)
• Add missing Detection struct (#27)
• Fix ops missing from upstream (#21)
• Fixes ORT segfault in nightly version (#18)
• Fix loading with coco and voc datasets (#12)
• Correcting incorrect types (#3)

Documents

• Update CODE_OF_CONDUCT.md (#61)
• Cleanup requirements.txt (#57)
• Update readme and setup instructions [skip ci] (#56, #19, #13)
• Add ultralytics like model loading notebooks (#52)
• Create CONTRIBUTING.md (#36) and CODE_OF_CONDUCT.md (#35) , thanks @BobinMathew !
• Fix badge links of GH actions (#32)
• Add statement of stable branch (#30)
• Make it more readable (#20)
• Update checkpoints updating information (#15)
• Add design principle (#14)
• Update Namespace arguments in updating checkpoints (#10)
• Use yolov5s, yolov5m and yolov5l directly (#9)

Model Checkpoints for r3.1 and r4.0

Last modified date: 2021-02-22

Just leave a block of space to store the model checkpoints.

NOTE: All checkpoints here make use of MD5 hash.

Support yolov5m and yolov5l models

New Features

• Support yolov5m and yolov5l models (#7)

Add graph visualization tools

New Features

• Add libtorch and onnx inference and export notebooks. Nov. 21, 2020.
• Add graph visualization tools and notebooks. Nov. 21, 2020.

Refactor the BackboneWithPAN module

New Features

• Refactor the Backbone modules, Nov. 16, 2020.
• Support exporting to onnx, and doing inference using onnxruntime. Nov. 17, 2020.

Refactor the YOLOHead and AnchorGenerator modules

New Features

• Add TorchScript cpp interface example, Nov. 4, 2020.
• Refactor the YoloHead and AnchorGenerator modules, Nov. 11, 2020.

yolort initial release

New Features

• Support exporting to TorchScript model, and doing inference using python interface, Oct. 8, 2020.
• Support doing inference using libtorch cpp interface, Oct. 10, 2020.