sujitojha1 / EVA4-S7

EVA4 Assignment S7, TSAI

S7

EVA4 Assignment S7, TSAI

Step 1

1. Setup the code and define the baseline architecture.
2. Set the device to GPU instead of CPU.
3. Exploratory data to understand the data normalization and data augmentation.

Results

1. Total parameters = 2,435,744
2. For 2 epochs, Train Accuracy 63%
3. For 2 epochs, Test Accuracy 62%

Analysis

1. Not able to achieve target accuracy of 80% and parameters less than 1M.

Step 2

1. Set the image normalization and image augmentation

Results .

1. Total parameters = 2,435,744
2. For 2 epochs, Train Accuracy 60%
3. For 2 epochs, Test Accuracy 62%

Analysis

1. Accuracy didn't improve as we tried only for two epochs.

Step 3

1. Refining the architecture to meet parameter target < 1M

Results

1. Total parameters = 1,092,768
2. For 20 epochs, Train Accuracy 84%
3. For 20 epochs, Test Accuracy 81%

Analysis

1. Accuracy meets the target >80% but slightly over the parameter targets 1M.

Steps 4

Implemented depth wise convolution and dilated layer. Even optimized parameter to meet target of 1M

Results

1. Total parameters = 943,552
2. For 20 epochs, Train Accuracy 82%
3. For 20 epochs, Test Accuracy 81%

Analysis

1. Results of accuracy is more than 80% and parameters target < 1M.
2. Models is performing well also as gap between train and test accuracy is small.