Research map / Vision-language and multimodal / Human and embodied modeling
TGIF: A New Dataset and Benchmark on Animated GIF Description
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace nearest-neighbor retrieval with sequence-to-sequence LSTM description | TGIF description · TGIF test | METEOR: 7.0 -> 16.7 | Helped |
| use FrameNet rather than WordNet semantic roles (ablation) | SMT · TGIF test | METEOR: 9.6 -> 14.1 | Helped |
| use a single random frame instead of sequence input (ablation) | LSTM description · TGIF test | METEOR: 16.7 -> 15.4 | Hurt |
| remove spatial cropping (ablation) | LSTM description · TGIF test | METEOR: 16.1 vs 16.1 | No clear effect |
| randomly crop a temporal subsequence (ablation) | LSTM description · TGIF test | METEOR: 16.1 vs 16.1 | No clear effect |
| use spatially consistent crops across a temporal subsequence (ablation) | LSTM description · TGIF test | METEOR: 16.1 -> 16.2 | No clear effect |
| randomly initialize and freeze CNN weights instead of using pretrained weights (ablation) | LSTM description · TGIF test | METEOR: 16.1 -> 13.6 | Hurt |
| fine-tune pretrained CNN weights (ablation) | LSTM description · TGIF test | METEOR: 16.1 -> 16.7 | Helped |
| increase training data from 20% to 80% (ablation) | S2VT · TGIF test | METEOR: 15.0 -> 16.1 | Helped |
| pretrain on TGIF then fine-tune on the target movie dataset | S2VT · M-VAD | METEOR: 4.99 -> 5.17 | Helped |
| pretrain on TGIF then fine-tune on the target movie dataset | S2VT · MPII-MD | METEOR: 5.35 -> 5.42 | Helped |
| pretrain on TGIF then fine-tune on the target movie dataset | S2VT · LSMDC | METEOR: 5.82 -> 5.77 | No clear effect |
Closest papers by meaning
- SEINE: Short-to-Long Video Diffusion Model for Generative Transition and Prediction (ICLR 2024 · similarity 0.82)
- Panda-70M: Captioning 70M Videos with Multiple Cross-Modality Teachers (CVPR 2024 · similarity 0.82)
- LLM4SGG: Large Language Models for Weakly Supervised Scene Graph Generation (CVPR 2024 · similarity 0.82)
- A Picture is Worth More Than 77 Text Tokens: Evaluating CLIP-Style Models on Dense Captions (CVPR 2024 · similarity 0.82)
- Distilling Vision-Language Models on Millions of Videos (CVPR 2024 · similarity 0.82)
- StoryGPT-V: Large Language Models as Consistent Story Visualizers (CVPR 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.