Training: some frames were used to learn the model. Validation: not used for gradient training, but used to select checkpoints. Unseen: absent from both saved lists; not necessarily untouched by later benchmarks. These are movie-level labels, not a claim that every screenshot was used in training.
Examples from the indexed movies. Save or copy an image to test it on the search page.
Groups refer to the movie’s training history, not whether each individual frame was used.
1 movies · 16 sample images
Chinatown [Unseen by training/validation] — frame_000921.jpgChinatown [Unseen by training/validation] — frame_001842.jpgChinatown [Unseen by training/validation] — frame_002763.jpgChinatown [Unseen by training/validation] — frame_003683.jpgChinatown [Unseen by training/validation] — frame_004604.jpgChinatown [Unseen by training/validation] — frame_005525.jpgChinatown [Unseen by training/validation] — frame_006445.jpgChinatown [Unseen by training/validation] — frame_007366.jpgChinatown [Unseen by training/validation] — frame_008287.jpgChinatown [Unseen by training/validation] — frame_009208.jpgChinatown [Unseen by training/validation] — frame_010128.jpgChinatown [Unseen by training/validation] — frame_011049.jpgChinatown [Unseen by training/validation] — frame_011970.jpgChinatown [Unseen by training/validation] — frame_012890.jpgChinatown [Unseen by training/validation] — frame_013811.jpgChinatown [Unseen by training/validation] — frame_014732.jpg