AV_GOT_DeepLearning

AV datahack on deep learning CV

The repo includes most models tried on AV Computer vision hackathon. Pointers:

Fastai significantly out performs pure pytorch based models
Around 30 different models were tried mainly from Cadene's pretrained repo, DenseNet, SENET, polynet,Inception variations,ResNets, Resnets with BAM, and some variations of pretrained EfficientNet
Focal loss works best for training instead of CrossEntropy loss

Final Standing: Public LB: 5th, Private LB: 7th

Name		Name	Last commit message	Last commit date
Latest commit History 8 Commits
AV_Datahack_v2.ipynb		AV_Datahack_v2.ipynb
AV_DeepLearning_Starter_1 (1).ipynb		AV_DeepLearning_Starter_1 (1).ipynb
AV_DeepLearning_Starter_1.ipynb		AV_DeepLearning_Starter_1.ipynb
AV_Densenet_FineTune.ipynb		AV_Densenet_FineTune.ipynb
AV_Fastai_ResNet_Variants.ipynb		AV_Fastai_ResNet_Variants.ipynb
AV_Fastai_v1.ipynb		AV_Fastai_v1.ipynb
AV_OtherArch_v1.ipynb		AV_OtherArch_v1.ipynb
Best_Single_Model_se_resnext101_32x4d.ipynb		Best_Single_Model_se_resnext101_32x4d.ipynb
DenseNet121_9711.ipynb		DenseNet121_9711.ipynb
DenseNet_9720.ipynb		DenseNet_9720.ipynb
Effnet_9728.ipynb		Effnet_9728.ipynb
Effnet_9784.ipynb		Effnet_9784.ipynb
Effnet_9812.ipynb		Effnet_9812.ipynb
Ensemble_WorkingFile_v1.xlsx		Ensemble_WorkingFile_v1.xlsx
InceptionV4.ipynb		InceptionV4.ipynb
Phaedrus2703_Approach.docx		Phaedrus2703_Approach.docx
README.md		README.md
SeResNext_9767.ipynb		SeResNext_9767.ipynb
Xception.ipynb		Xception.ipynb
av-fastai-v1.ipynb		av-fastai-v1.ipynb
av-resnet-d-101b.ipynb		av-resnet-d-101b.ipynb
bottleneckmodels-av-v1-0.ipynb		bottleneckmodels-av-v1-0.ipynb
readme.txt		readme.txt

Provide feedback