Reproduce “General Multi-label Image Classification with Transformers” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jack Lanchantin, Tianlu Wang, Vicente Ordonez, Yanjun Qi
C-Tran completes missing labels in multi-label images by combining image features with known positive, negative, or unknown states. It trains with masks so the model predicts labels left hidden, using a Transformer to process image and label embeddings. On COCO-80 with 75% of labels supplied, it reaches 86.0% mAP on unknown labels versus 80.9% for Feedback-prop; labels are simulated and come from its training vocabulary.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.