Research map / Vision-language and multimodal / Human and embodied modeling
BAM! The Behance Artistic Media Dataset for Recognition Beyond Photography
Michael J. Wilber, Chen Fang, Hailin Jin, Aaron Hertzmann, John Collomosse, Serge Belongie
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| train on BAM art labels instead of VOC/ImageNet | object classifier · BAM VOC categories average | AP: 0.9448 vs 0.8770 (YOLO), 0.8801 (SSD), 0.8567 (RN50) | Improved |
| fuse ImageNet and BAM scores linearly | fine-tuned ResNet-50 · BAM VOC categories average | AP: 0.9448 -> 0.9512 | Improved |
| train jointly on BAM attributes (JAM) | ResNet-50 · Flickr style | accuracy: 0.376 -> 0.389 | Improved |
| train jointly on BAM attributes (JAM) | ResNet-50 · Wikipaintings style | accuracy: 0.505 -> 0.508 | No clear effect |
| train jointly on BAM attributes (JAM) | ResNet-50 · AVA style | accuracy: 0.603 -> 0.615 | Improved |
| train on seen media instead of unseen medium | object classifier · Cross-medium object recognition (watercolor) | accuracy: 63 -> 68 | Improved |
| train on 6 other media instead of ImageNet only | object classifier · Test on 3D | accuracy: 52 -> 70 | Improved |
| use StyleNet features instead of ImageNet ResNet features | linear SVM · BAM emotion and media attributes | AP: 4/4 emotion and 4/6 media attributes | Improved |
| use StyleNet features instead of ImageNet ResNet features | linear SVM · BAM content attributes | AP: higher AP for ImageNet on all nine content attributes | Hurt |
| use task-specific subspaces instead of shared features | retrieval embedding · BAM retrieval precision-at-10 (media) | precision-at-10: 0.80 -> 0.91 | Improved |
| use task-specific subspaces instead of shared features | retrieval embedding · BAM retrieval precision-at-10 (emotion) | precision-at-10: 0.80 -> 0.84 | Improved |
| use task-specific subspaces instead of shared features | retrieval embedding · BAM retrieval precision-at-10 (content) | precision-at-10: 0.80 -> 0.71 | Hurt |
Closest papers by meaning
- APDDv2: Aesthetics of Paintings and Drawings Dataset with Artist Labeled Scores and Comments (NeurIPS 2024 · similarity 0.85)
- Stylebreeder: Exploring and Democratizing Artistic Styles through Text-to-Image Models (NeurIPS 2024 · similarity 0.84)
- VADB: A Large-Scale Video Aesthetic Database with Professional and Multi-Dimensional Annotations (NeurIPS 2025 · similarity 0.84)
- ArtBank: Artistic Style Transfer with Pre-trained Diffusion Model and Implicit Style Prompt Bank (AAAI 2024 · similarity 0.84)
- Gender Artifacts from Art History to Text-to-Image Generation (NeurIPS 2026 · similarity 0.84)
- Can Machines Understand Composition? Dataset and Benchmark for Photographic Image Composition Embedding and Understanding (CVPR 2025 · similarity 0.84)
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.
