PyTorch Sanity Check

---

Work in progress...

Data is assumed to be the iNaturalist dataset for the convolutional benchmark.

Most of the code here follows PyTorch examples.

Give it a go:

For dev purposes:

python inaturalist.py ./data --workers 8 -b 12 --gpu 0

For filling up four RTX 2080Ti and training with torch.DataParallel:

python inaturalist.py ./data -a resnet50 --workers 32 -b 192

Todo:

• Write a proper Readme.
• Add dependencies.
• Read from yaml config files the standardized benchmarks.
• Start adding additional models.