ML-Collections/Transfer-learning at main · hwaseem04/ML-Collections

Name		Name	Last commit message	Last commit date
parent directory ..
tiny_FR		tiny_FR
ConvNeXt.ipynb		ConvNeXt.ipynb
Inception-v3.ipynb		Inception-v3.ipynb
README.md		README.md
Vanilla-grade-sequential-CNN.ipynb		Vanilla-grade-sequential-CNN.ipynb
Vanilla-sequential-CNN.ipynb		Vanilla-sequential-CNN.ipynb

README.md

I have implemented 3 architectures. Two of which are fine-tuned on pre-trained architectures - Inception-V3 and ConvNext and one of which is a vanilla grade architecture.
Stored best epoch checkpoints for inference.
Used custom dataset class to retrieve data
- Need to properly apply transformations to train and test data. As of now I haven't done data augmentation. Need to do it and check performance.
Inception-V3 Implementation
- F1 score ~ 0.80
ConvNeXt Implementation
- F1 score ~ 0.82
Vanilla Grade sequential CNN Implementation
- F1 score ~ 0.52
- It is very poor, need to iterate on the architecture. Tried Ray Tune, but didn't configure it properly. Need to do that.