pyronear / pyro-vision

Computer vision library for wildfire detection 🌲 Deep learning models in PyTorch & ONNX for inference on edge devices (e.g. Raspberry Pi)
https://pyronear.org/pyro-vision/
Apache License 2.0
51 stars 24 forks source link

refactor: Refactored package for incoming v0.2 #138

Closed frgfm closed 2 years ago

frgfm commented 2 years ago

This PR introduces the following modifications:

The goal of this PR is to help speed up iterations by R&D teams. PyroVision is not meant primarily for model exploration, Pyronear curates and provides relevant data, and model that will be trained on them. Thus features for model modification have been deprecated. Next PR will focus on checkpoint sharing and loading via HF Hub!

Any feedback is welcome!

codecov[bot] commented 2 years ago

Codecov Report

Merging #138 (244952c) into master (40ddf2f) will decrease coverage by 3.16%. The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #138      +/-   ##
==========================================
- Coverage   92.33%   89.16%   -3.17%     
==========================================
  Files          18        8      -10     
  Lines         652      203     -449     
==========================================
- Hits          602      181     -421     
+ Misses         50       22      -28     
Flag Coverage Δ
unittests 89.16% <100.00%> (-3.17%) :arrow_down:

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files Coverage Δ
pyrovision/datasets/__init__.py 100.00% <ø> (ø)
pyrovision/datasets/openfire.py 91.56% <100.00%> (+0.54%) :arrow_up:
pyrovision/datasets/utils.py 77.61% <100.00%> (-13.16%) :arrow_down:
pyrovision/models/__init__.py 100.00% <100.00%> (ø)
pyrovision/models/mobilenetv3.py 100.00% <100.00%> (ø)
pyrovision/models/resnet.py 100.00% <100.00%> (+17.39%) :arrow_up:
pyrovision/models/rexnet.py 100.00% <100.00%> (+4.16%) :arrow_up:

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update ce4db67...244952c. Read the comment docs.

MateoLostanlen commented 2 years ago

our current model used by pyro-engine was based on this architecture it not possible to load the model anymore now :/

frgfm commented 2 years ago

Yes, that's why the release is on hold while we haven't published new sets of params 👍 Sorry about the incovenience!

To use the previous one, one can checkout the commit right before this one. But the goal is to make everything clean for 0.2.0. I still need to retrain our baseline on those architectures and publish it

If you have training undergoing, I'd suggest using the updated scripts & architectures (should get better perf) 👍