Research map / Image generation / Image restoration
Discovering Adaptive Task Dependencies for Efficient Multi-Task Representation Compression
Zhimeng Huang, Rongao Yuan, Junlong Gao, Qi Mao, Siwei Ma, Wen Gao, Chuanmin Jia
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 |
|---|---|---|---|
| use adaptive DAG task order instead of fixed order (ablation) | ATDC · Taskonomy Tiny, semantic segmentation and keypoint 2D | task loss across bitrates | Improved |
| use fixed task order instead of random per-image order (ablation) | ATDC · Taskonomy Tiny, semantic segmentation and keypoint 2D | task loss across bitrates | Improved |
| add an unseen keypoint 2D task by fine-tuning only the proxy head and a new predictive module (ablation) | ATDC trained on four tasks · Taskonomy Tiny, tasks s and n | task loss: 0.1837 -> 0.1841 (s); 0.2115 -> 0.2110 (n) | No clear effect |
| add per-image DAG construction versus a fixed order | ATDC · Taskonomy Tiny, encoding latency | encoding time (ms): 41.3 -> 44.8 | Hurt |
| use ATDC instead of TAMC on edge texture and surface normal | multi-task compression · Taskonomy Tiny | task accuracy | Improved |
| use ATDC instead of TAMC for image reconstruction | multi-task compression · Taskonomy Tiny, reconstruction | PSNR, FID, LPIPS, MS-SSIM | Mixed |
| use TAMC shared first-stage representation versus ATDC sequential decoding | scalable multi-task coding · Taskonomy Tiny, cumulative utility vs bitrate | cumulative normalized loss | Mixed |
Closest papers by meaning
- Test-Time Fine-Tuning of Image Compression Models for Multi-Task Adaptability (CVPR 2025 · similarity 0.85)
- When MLLMs Meets Compression Distortion: A Coding Paradigm Tailored to MLLMs (ICLR 2026 · similarity 0.85)
- Towards Unified Human Perception and Machine Understanding: Token Flow Guided Compression Framework (CVPR 2026 · similarity 0.84)
- Prune Redundancy, Preserve Essence: Vision Token Compression in VLMs via Synergistic Importance-Diversity (ICLR 2026 · similarity 0.84)
- Going Beyond Multi-Task Dense Prediction with Synergy Embedding Models (CVPR 2024 · similarity 0.84)
- All-in-One Image Coding for Joint Human-Machine Vision with Multi-Path Aggregation (NeurIPS 2024 · 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.
