facebookresearch/deepcluster

Health v2 · maintenance only

Flags: no_releases archived no_license

Adoption not part of the score

1743 stars · 318 forks observed · 2026-08-28

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

DeepCluster is a research codebase from Facebook AI Research implementing the 'Deep Clustering for Unsupervised Learning of Visual Features' method, which trains convolutional neural networks without labels by iteratively clustering features and using cluster assignments as pseudo-labels. It includes training code, evaluation protocols (Pascal VOC classification, linear probing, image retrieval), a feature visualization module, and pre-trained AlexNet and VGG-16 models.

Use cases

• train convnets without labeled data using deep clustering
• reproduce DeepCluster results on ImageNet
• evaluate unsupervised visual features with linear classification
• perform instance-level image retrieval with learned features
• classify Pascal VOC images with self-supervised features
• visualize quality of learned visual representations
• download pre-trained AlexNet and VGG-16 self-supervised models

When to choose

• you need the original DeepCluster method for unsupervised feature learning research
• you want pre-trained self-supervised AlexNet/VGG-16 backbones
• you need baseline evaluation protocols for unsupervised visual representations

When to avoid

• you want the improved DeepCluster-v2 or SwAV methods - use the swav repository instead
• you need modern PyTorch versions - the code targets PyTorch 0.1.8 and Python 2.7
• you need production-ready or maintained software rather than research code

Facets

library · maturity maintenance

machine-learning deep-learning image-processing computer-vision machine-learning deep-learning python self-supervised-learning unsupervised-learning clustering pytorch convnets research-code pretrained-models imagenet gpu linux

Member repositories

For agents

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