Thanks for sharing. Nice approach using progressive resizing. I had used fastai2 and kfold validation using xresnet50. Got 14th place. Will post code soon. Had faced some reproducibility issue which was reduced by cross validation. Could you please tell if you faced any reproducibility issue by using. Learnt a lot from Jonathan Whitaker and Radek Osmulski

Hi aninda_bitm,

Yes, I did have reproducibility issues. After a quick search, I have found the following code which seems to work (at least until you restart your kernel).

https://forums.fast.ai/t/solved-reproducibility-where-is-the-randomness-coming-in/31628/11

https://docs.fast.ai/dev/test.html#getting-reproducible-results

Hope this helps:

seed = 42

# python RNG

import random

random.seed(seed)

# pytorch RNGs

import torch

torch.manual_seed(seed)

torch.backends.cudnn.deterministic = True

if torch.cuda.is_available(): torch.cuda.manual_seed_all(seed)

# numpy RNG

import numpy as np

np.random.seed(seed)

hi ,

thanks for sharing your solution and congratulations ,I tried to use the pre trained model DENSNET161 but there is a problem

problem is : CUDA out of memory. Tried to allocate 2.00 MiB (GPU 0; 7.43 GiB total capacity; 6.85 GiB already allocated; 2.94 MiB free; 6.91 GiB reserved in total by PyTorch)