facebookresearch/FixRes

Health v2 · maintenance only

Flags: no_releases archived no_license

Adoption not part of the score

1043 stars · 146 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

FixRes is a PyTorch implementation of the NeurIPS 2019 paper 'Fixing the train-test resolution discrepancy', providing training and fine-tuning code plus pre-trained ImageNet classification models. It improves accuracy of convolutional neural networks by adapting models to a higher test resolution than used during training.

Use cases

• improve image classification accuracy with higher test resolution
• fine-tune a pretrained ResNet for ImageNet at higher resolution
• download pretrained FixResNet models for image classification
• reproduce the FixRes paper results
• apply train-test resolution adaptation to my own CNN

When to choose

• you want a simple accuracy boost for CNN image classifiers without architecture changes
• you need pretrained ImageNet models at higher resolutions like 320-384px
• you are reproducing or building on the FixRes research

When to avoid

• you need actively maintained code with modern PyTorch support
• you want a production training framework rather than research code
• you work with transformers rather than convolutional networks
• you need convenient multi-GPU distributed training out of the box

Facets

library · maturity maintenance

machine-learning deep-learning image-processing computer-vision image-processing deep-learning machine-learning python pytorch image-classification imagenet train-test-resolution research-code pretrained-models fine-tuning gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="facebookresearch/FixRes")