Research map / Vision-language and multimodal / Human and embodied modeling
Uni-Perceiver v2: A Generalist Model for Large-Scale Vision and Vision-Language Tasks
Read the paperCheck 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 |
|---|---|---|---|
| gather retrieval features across GPUs (larger negative batch) under unmixed sampling (ablation) | Uni-Perceiver v2 (ResNet-50) · COCO retrieval | I2T R@1: 39.8 -> 50.4 | Helped |
| add MT-AdamW optimizer over unmixed sampling with gather (ablation) | Uni-Perceiver v2 (ResNet-50) · COCO caption | BLEU-4: 27.6 -> 30.6 | Helped |
| add MT-AdamW optimizer (ablation) | Uni-Perceiver v2 (ResNet-50) · COCO retrieval | I2T R@1: 50.4 -> 51.3 | Helped |
| use unmixed instead of mixed task sampling (no gather) (ablation) | Uni-Perceiver v2 (ResNet-50) · COCO detection | mAP: 49.6 -> 49.2 | Hurt |
| add global representation to regional representation (ablation) | Uni-Perceiver v2 (ResNet-50) · COCO retrieval | I2T R@1: 52.3 -> 51.3 | Hurt |
| add global representation to regional representation (ablation) | Uni-Perceiver v2 (ResNet-50) · ImageNet-1k | accuracy: 75.9 -> 76.9 | Helped |
| add regional representation to global-only (ablation) | Uni-Perceiver v2 (ResNet-50) · COCO caption | BLEU-4: 28.8 -> 30.6 | Helped |
| add Conditional MoE to multi-task training (ablation) | Uni-Perceiver v2 (ResNet-50) · COCO retrieval | I2T R@1: 46.0 -> 51.3 | Helped |
| joint multi-task training vs single task (no MoE) (ablation) | Uni-Perceiver v2 (ResNet-50) · COCO retrieval | I2T R@1: 50.0 -> 46.0 | Hurt |
| remove classification task from joint training (ablation) | Uni-Perceiver v2 (ResNet-50) · COCO caption | BLEU-4: 28.9 -> 25.9 | Hurt |
| use ImageNet-21k instead of ImageNet-1k pretrained image encoder (ablation) | Uni-Perceiver v2 (ResNet-50) · ImageNet-1k | accuracy: 76.8 -> 80.1 | Helped |
| use CLIP instead of supervised IN-1k pretrained encoder (ablation) | Uni-Perceiver v2 (ResNet-50) · COCO detection | mAP: 49.9 -> 47.2 | Hurt |
Closest papers by meaning
- VisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Tasks (NeurIPS 2024 · similarity 0.89)
- DICEPTION: A Generalist Diffusion Model for Visual Perceptual Tasks (NeurIPS 2025 · similarity 0.89)
- Perception Encoder: The best visual embeddings are not at the output of the network (NeurIPS 2025 · similarity 0.88)
- Enhancing Target-unspecific Tasks through a Features Matrix (ICML 2025 · similarity 0.88)
- Large Vision-Language Models Get Lost in Attention (ICML 2026 · similarity 0.88)
- Benchmarking Large Vision-Language Models on Fine-Grained Image Tasks: A Comprehensive Evaluation (ICLR 2026 · similarity 0.88)
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.