sumitsarkar1 / assignment9

Create a Custom Resnet Model and train for 90% Test accuracy

Create a Custom Resnet Model and train for 90% Test accuracy on Cifar10 Dataset

Design Constraints

1. Model to be used : custom Resnet18

2. Number of epochs to be used: 24

3. Learning Rate (LR) Policy : One Cycle LR with max LR at 5th epoch

4. Data Augmentation to be used : Random Crop, Horizontal Flip, CutOut

4. Test Accuracy to achieve : 90%

Final Test Accuracy achieved : 93%+

Training/Testing Loss and Accuracy , Variation of LR with epochs

Missclassified Images and their gradcam analysis

Last 4 epoch results

EPOCH: 21
Loss=0.15072710812091827 Batch_id=97 Train Accuracy=93.65: 100%|█| 98/98 [00:22<
Test set: Average loss: 0.0005, Accuracy: 9315/10000 (93.15%)

EPOCH: 22
Loss=0.12879112362861633 Batch_id=97 Train Accuracy=93.87: 100%|█| 98/98 [00:23<
Test set: Average loss: 0.0005, Accuracy: 9335/10000 (93.35%)

EPOCH: 23
Loss=0.1782904863357544 Batch_id=97 Train Accuracy=94.07: 100%|█| 98/98 [00:23<0
Test set: Average loss: 0.0004, Accuracy: 9327/10000 (93.27%)

EPOCH: 24
Loss=0.1502068042755127 Batch_id=97 Train Accuracy=94.32: 100%|█| 98/98 [00:23<0
Test set: Average loss: 0.0004, Accuracy: 9352/10000 (93.52%)