The largest collection of PyTorch image encoders / backbones. Including train, eval, inference, export scripts, and pretrained weights -- ResNet, ResNeXT, EfficientNet, NFNet, Vision Transformer (ViT), MobileNetV4, MobileNet-V3 & V2, RegNet, DPN, CSPNet, Swin Transformer, MaxViT, CoAtNet, ConvNeXt, and more
validate.py: add --seed for reproducible validation
Mirrors what train.py already does: - new --seed CLI arg (default 42, matching train.py) - utils.random_seed(args.seed) early in validate() - seed=args.seed plumbed into create_dataset() Without this, validate.py has no way to fix Python/NumPy/PyTorch RNG state or the dataset reader's shuffle seed, so repeated runs against the same checkpoint can produce slightly different results when the dataset reader is stochastic (TFDS subsplits, IterableDatasets, partial-sample validation, etc.). No behavior change for users who don't pass --seed.
J
Jan Bernloehr committed
d45cb111f0f5c575c1ed6f92e346985908002340
Parent: 0e968e1
Committed by Ross Wightman <rwightman@users.noreply.github.com>
on 5/21/2026, 4:45:20 AM