Add sub sampler resnet18

[MateoLostanlen]

The goal of this PR is to add the SubSamplerResnet18 model which is a derivative of the resnet18 that we want to train on the SubSamplerDataSet of the PR #66

[codecov]

Codecov Report

Merging #69 into master will increase coverage by 0.53%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master      #69      +/-   ##
==========================================
+ Coverage   85.21%   85.74%   +0.53%     
==========================================
  Files          20       21       +1     
  Lines         771      821      +50     
==========================================
+ Hits          657      704      +47     
- Misses        114      117       +3     

Flag	Coverage Δ
#unittests	85.74% <100.00%> (+0.53%)	⬆️

Impacted Files	Coverage Δ
pyronear/models/__init__.py	100.00% <100.00%> (ø)
pyronear/models/ssresnet.py	100.00% <100.00%> (ø)
pyronear/utils/collect_env.py	70.66% <0.00%> (-1.34%)	⬇️
pyronear/datasets/wildfire/split_strategy.py	97.29% <0.00%> (+0.52%)	⬆️

---

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 61e6410...e4006c3. Read the comment docs.

[frgfm]

Great PR! I made a few suggestions to get this to be easily reusable on other resnet-based achitecture, let me know what you think!

[MateoLostanlen]

Hi @frgfm , Thanks for your remarks. It is indeed a good idea to generalize to any resnet. Let me know what you think of this new version

[frgfm]

Almost there! you can define each ssresnet size like in torchvision and move weight loading outside of the class definition, check my comments, and we'll be good to go!

[frgfm]

Thanks for the PR! This already looks good, I made a few suggestions to improve flexibility, let me know your thoughts on it!

[frgfm]

A few minor adjustments for the final touch, and we're good to go!

[frgfm]

Thanks a lot, looking forward to try it out!