[Food Vision Milestone] Model takes a long time to train

[ishan-pathirana]

I am working on the milestone project and my code seems to take a long time to train even with mixed precision.

• Downgraded the TensorFlow version to 2.4.1 to enable mixed precision on EfficientNetB0
• Tried the notebook in the repo as well (with downgraded TensorFlow 2.4.1)
• It doesn't reduce ETA even after 15 mins

Saving TensorBoard log files to: training_logs/efficientnetb0_101_classes_all_data_featrue_extract/20221020-133511 Epoch 1/3 4/2368 [..............................] - ETA: 94:45:51 - loss: nan - accuracy: 0.0254

[ishan-pathirana]

I have replaced EfficientNetB0 with ResNet50 and it solved the training time issue. Also, no need to downgrade TensorFlow to run ResNet50.

[ishandandekar]

Hi @ishan-anuranga , did you try modifying the data to user more processing power? I think @mrdbourke , used prefetching and batching the data for faster loading time.