ViRA: Toward Ultra-High-Resolution Visual Agents via Cross-Input Relative Policy Optimization
Organizations: Rochester Institute of Technology · Purdue University · University of Utah · NVIDIA
Abstract
Visual agents answer questions about ultra-high-resolution photos by zooming into the relevant regions. However, existing agents often answer from common sense instead of reading the evidence. Reinforcement learning (RL) on the final answer cannot correct this, because a guess earns the same reward as reading. We propose Cross-Input Relative Policy Optimization (CrossRPO), which trains agents on contrast sets. A contrast set asks one question about several versions of an image that differ only in a small region with the evidence and have different answers, so only reading the evidence solves the whole set. CrossRPO rewards solving the whole set, and we prove that it does not reinforce a set answered from common sense. We build contrast sets for training (ViRA-Train) and evaluation (ViRA-Bench) by editing one small object in images of 21 to 120 megapixels, and we photograph further sets with real changes (ViRA-Bench-Real) to exclude editing traces. We train two Visual Reasoning Agents, ViRA-9B and ViRA-27B, with CrossRPO on only about 3,500 items of ViRA-Train each. CrossRPO achieves 63.3% set accuracy on ViRA-Bench at the maximum resolution, against 59.3% for GRPO. ViRA-9B is the most accurate open model of its size on all benchmarks, and ViRA-27B outperforms Claude Haiku 5.5 and GPT-6-Luna, both at their highest thinking level, on most benchmarks.
Figures & tables
| Max Res. | ViRA-Bench | ViRA-Bench-Real | |||||||||
| Model | (MP) | 0.5 MP | 1 MP | 2 MP | 4 MP | Max | 0.5 MP | 1 MP | 2 MP | 4 MP | Max |
| Open 7–9B models | |||||||||||
| Qwen3.5-9B | 16.8 | 0.9 (2K) | 2.2 (4K) | 3.8 (6K) | 7.0 (12K) | 12.2 (49K) | 0.0 (2K) | 0.0 (3K) | 0.0 (6K) | 2.6 (12K) | 7.7 (49K) |
| ZwZ2-9B | 16.8 | 2.1 (4K) | 3.8 (5K) | 9.2 (8K) | 13.9 (14K) | 24.1 (51K) | 0.0 (3K) | 0.0 (5K) | 0.0 (8K) | 2.6 (14K) | 17.9 (50K) |
| Vision-OPD-9B | 16.8 | 2.1 (2K) | 5.1 (3K) | 8.0 (6K) | 12.3 (12K) | 20.5 (49K) | 0.0 (2K) | 0.0 (3K) | 1.3 (6K) | 2.6 (12K) | 24.4 (49K) |
| Open 7–9B models + tool | |||||||||||
| V ∗ | HR-Bench 4K | HR-Bench 8K | MME-RW-Lite | TreeBench | O3-Bench | MMStar | Hallusion | |||||||
| Model | 1 MP | Max | 1 MP | Max | 1 MP | Max | 1 MP | Max | 1 MP | Max | 1 MP | Max | Max | Max |
| Open 7–9B models | ||||||||||||||
| Qwen3.5-9B | 74.9 | 82.2 | 64.9 | 83.2 | 60.2 | 79.2 | 52.8 | 55.5 | 44.2 | 46.7 | 27.2 | 53.3 | 70.8 | 74.1 |
| ZwZ2-9B | 82.2 | 90.1 | 72.9 | 87.9 | 65.4 | 84.8 | 57.1 | 61.3 | 49.6 | 48.4 | 26.1 | 59.4 | 76.3 | 78.5 |
| Vision-OPD-9B | 80.6 | 94.2 | 67.9 | 83.0 | 63.2 | 80.9 | 57.6 | 61.5 | 46.7 | 47.7 | 30.1 | 53.6 | 72.7 | 74.1 |
| Open 7–9B models + tool | ||||||||||||||
| ViRA-Bench | ViRA-Bench-Real | V ∗ | HR-4K | HR-8K | MME-RW | TreeBench | O3-Bench | MMStar | Hallusion | |||||||||||||||
| Advantage | 0.5 | 1 | 2 | 4 | Max | 0.5 | 1 | 2 | 4 | Max | 1 | Max | 1 | Max | 1 | Max | 1 | Max | 1 | Max | 1 | Max | Max | Max |
| CrossRPO ( ) | 38.7 | 46.9 | 55.6 | 57.5 | 63.3 | 31.2 | 49.6 | 57.7 | 68.8 | 77.4 | 92.0 | 96.2 | 87.5 | 91.0 | 84.7 | 90.1 | 63.6 | 65.3 | 56.3 | 57.0 | 50.4 | 67.0 | 77.1 | 79.6 |
| Joint term | ||||||||||||||||||||||||
| (per-version RLOO) | 31.3 | 41.2 | 50.0 | 52.9 | 58.7 | 23.5 | 41.9 | 56.8 | 65.4 | 70.1 | 93.0 | 95.6 | 87.8 | 92.9 | 85.7 | 89.8 | 62.4 | 64.1 | 54.3 | 55.3 | 43.5 | 64.1 | 78.8 | 80.2 |
| 33.4 | 44.2 | 51.3 | 56.1 | 56.6 | 32.9 | 49.6 | 59.0 | 67.1 | 69.2 | 89.5 | 94.6 | 85.4 | 89.8 | 83.4 | 89.1 | 60.2 | 64.0 | 52.8 | 53.6 | 47.8 | 62.3 | 77.4 | 78.3 | |
| 40.9 | 47.9 | 53.8 | 55.7 | 58.2 | 52.1 | 59.0 | 68.8 | 69.7 | 74.8 | 87.3 | 93.9 | 82.4 | 88.9 | 81.8 | 86.8 | 60.3 | 62.6 | 52.3 | 53.1 | 40.3 | 58.3 | 76.5 | 77.0 | |
| Base | Pooled GRPO | Pooled RLOO | CrossRPO | ||
|---|---|---|---|---|---|
| ConFiQA, all three correct | 58.6 | 81.0 1.3 | 79.6 1.8 | 80.2 1.7 | 81.9 3.1 |
| strong-prior questions (493) | 64.3 | 90.5 0.4 | 91.5 3.5 | 90.3 1.0 | 91.5 2.0 |
| O correct | 76.0 | 87.1 1.7 | 86.4 2.2 | 87.0 1.6 | 87.8 2.3 |
| A correct | 75.7 | 90.6 1.2 | 90.5 1.1 | 90.0 1.2 | 91.1 1.5 |
| B correct | 71.8 | 90.6 1.5 | 91.0 0.4 | 89.4 1.7 | 91.0 1.6 |
| A or B with the original fact | 2.4 | 0.1 0.1 | 0.1 0.1 | 0.2 0.1 | 0.0 0.0 |
| All versions correct | Accuracy per version | |||||||
|---|---|---|---|---|---|---|---|---|
| Pooled GRPO | Pooled RLOO | CrossRPO | Pooled GRPO | Pooled RLOO | CrossRPO | |||
| 3 | 81.0 1.3 | 79.6 1.8 | 80.2 1.7 | 81.9 3.1 | 89.4 1.1 | 89.3 1.2 | 88.8 1.4 | 90.0 1.8 |
| 4 | 79.2 3.0 | 81.2 2.4 | 79.0 1.2 | 82.0 3.1 | 89.7 1.3 | 90.7 1.0 | 89.4 0.7 | 91.5 1.0 |
| 8 | 73.4 3.8 | 79.7 0.7 | 76.8 4.2 | 81.4 3.5 | 91.0 1.4 | 92.6 0.8 | 91.4 0.9 | 93.4 1.3 |
| 16 | 77.9 1.7 | 76.5 2.3 | 78.7 5.6 | 78.9 1.0 | 93.2 1.1 | 93.0 1.4 | 94.5 0.9 | 94.6 0.8 |
Appendix figures & tables27 assets
Supplementary material from the paper’s appendix.
Appendix
| Method | Answers | Joint term (D1) | Baseline (D2) | Scale (D3) | ||
|---|---|---|---|---|---|---|
| RLOO [ 2 ] | 1 | – | 0 | – | leave-one-out | none |
| Dr. GRPO [ 44 ] | 1 | – | 0 | – | group mean | none |
| GRPO [ 55 ] | 1 | – | 0 | – | group mean | per group |
| NoisyRollout † [ 43 ] | 1 | same | 0 | – | pooled with noisy copies | per group |
| COMET [ 89 ] | 2 | same | 0 | – | pooled over versions | per group |
| DNA-Train [ 27 ] | 2 | differ | 0 | – | per version | per version, |
| Requirement | CrossRPO | Alternative in prior work | What the alternative loses |
|---|---|---|---|
| Target | set accuracy for , all-correct triples for | per-version accuracy; agreement | a blind policy reaches on one version, and agreement is maximized by blind policies; the set accuracy loses its weight as grows (Propositions A.11 and A.12 , Corollary A.31 and Proposition A.32 ) |
| Unbiasedness | leave-one-out within each version | division by group statistics | a bias from statistics that contain the rollout itself [ 44 ] ; CrossRPO is unbiased for every (Theorem A.19 ) |
| Variance (D1) | all combinations | pairing by index | covariance larger by ; first-order variance ratio as grows (Theorem A.23 and Corollary A.24 ) |
| Variance (D2) | per-version baseline | baseline pooled over versions | an extra from the difficulty gap between versions (Proposition A.25 ) |
| Credit (D2) | per-version baseline and | baseline pooled over versions | answers from the prior get instead of 0, and credit shifts toward easier versions (Propositions A.26 and A.27 ) |
| Scale (D3) | one statistic per batch | standard deviation per group or set | a cost difference of any size becomes the advantage (Proposition A.29 ) |
| Members of a set | Shared by the members | Shortcut whose set accuracy is bounded | Effect of the joint term |
|---|---|---|---|
| Versions of one photo that differ in the evidence (contrast sets) | scene outside , question | answering from the scene and common sense without inspecting the region | bounds the shortcut by a fraction (Proposition A.40 (a)) |
| Different photos with the same question and different answers [ 17 ] | question | answering from the question alone | bounds the shortcut by a fraction (Proposition A.40 (a)) |
| Transformed copies with the same answer | everything but the transform | none: the answers are not distinct | rewards answering every copy correctly, but a shortcut can do so |
| Independent items | nothing | none (Proposition A.40 (b)) | in expectation a constant reweighting of RLOO, plus variance (Proposition A.38 ) |
| Set accuracy | 0.5 MP | 1 MP | 2 MP | 4 MP | Max | |
|---|---|---|---|---|---|---|
| ViRA-Bench | 0.5 | 38.7 | 46.9 | 55.6 | 57.5 | 63.3 |
| 0 | 31.3 | 41.2 | 50.0 | 52.9 | 58.7 | |
| ViRA-Bench-Real | 0.5 | 31.2 | 49.6 | 57.7 | 68.8 | 77.4 |
| 0 | 23.5 | 41.9 | 56.8 | 65.4 | 70.1 |
| Step | Model | Sets |
|---|---|---|
| Propose the object | Claude Sonnet 5.5 | 480 |
| Gemini 3.1 Pro | 49 | |
| GPT-6.1-Terra | 233 | |
| Write the question and the edits | Claude Opus 5.5 | 529 |
| GPT-6-Astra | 233 | |
| Edit the crop | GPT-5.4 Image 2 | 480 |
| ViRA-Bench | Real | ViRA-Train | |
|---|---|---|---|
| Contrast sets (questions) | 762 | 78 | 6,457 |
| Image sets | 544 | 78 | 4,462 |
| Photos | 466 | 234 | 3,797 |
| Photo size (MP) | 24–108 | 45.4 | 21–120 |
| Versions per set | 3 | 3 | 2–3 |
| Region that differs | 0.010% | – | 0.022% |
| 0.5 MP | 1 MP | 2 MP | 4 MP | Max | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Agent | O | A/B O | O | A/B O | O | A/B O | O | A/B O | O | A/B O |
| Qwen3.5-9B + tool | 80.1 | 0.3 | 80.6 | 0.3 | 84.0 | 0.3 | 86.5 | 0.1 | 85.7 | +0.5 |
| ZwZ2-9B + tool | 77.8 | 0.5 | 78.7 | +0.4 | 83.1 | 0.1 | 85.0 | +1.2 | 85.8 | 0.1 |
| 86.5 | +0.7 | 89.6 | +0.3 | 91.2 | 0.3 | 92.9 | 0.1 | 94.3 | +0.4 | |
| ViRA-9B | 87.6 | 0.4 | 89.9 | +0.4 | 92.4 | 0.0 | 94.2 | 0.1 | 96.2 | +0.2 |
| Qwen3.8-27B + tool | 75.5 | 1.1 | 81.5 | 0.3 | 86.9 | 0.3 | 89.5 | 0.3 | 90.6 | +0.1 |
| Source | ViRA-9B | ViRA-27B | Answer | |
|---|---|---|---|---|
| Contrast sets | 2–3 | 1,664 | 1,338 | multiple choice |
| Commons detail questions | 1 | 599 | 600 | open |
| Multi-evidence, photos | 1 | 859 | 949 | choice or open |
| Multi-evidence, maps | 1 | 170 | 175 | choice or open |
| Synthetic charts (5 types) | 1 | 200 | 200 | open |
| Math, ThinkLite-VL [ 68 ] | 1 | 150 | 139 | open |
| Setting | Value |
|---|---|
| Base model | Qwen3.5-9B / Qwen3.8-27B |
| Thinking | off |
| Framework | verl 0.10 [ 58 ] , vLLM 0.32 rollouts |
| Hardware | 8 H200 GPUs |
| Training time | 12.7 h (9B) / 29.6 h (27B) |
| Steps | 227 (9B) / 212 (27B) |
| Model | Checkpoint or API model | Training data |
| Ours | ||
| ViRA-9B | trained from Qwen/Qwen3.5-9B | 3,642 |
| ViRA-27B | trained from Qwen/Qwen3.8-27B | 3,401 |
| Open models | ||
| Qwen3.5-9B | Qwen/Qwen3.5-9B | – |
| ZwZ2-9B | inclusionAI/ZwZ2-9B | 20.4K |
| ViRA-Bench | ViRA-Bench-Real | |||||||||
| Model | 0.5 MP | 1 MP | 2 MP | 4 MP | Max | 0.5 MP | 1 MP | 2 MP | 4 MP | Max |
| Open 7–9B models | ||||||||||
| Qwen3.5-9B | 31.3 | 33.8 | 37.0 | 39.9 | 46.2 | 28.6 | 29.5 | 32.5 | 35.0 | 45.7 |
| ZwZ2-9B | 30.7 | 34.3 | 39.6 | 44.2 | 54.3 | 27.8 | 27.8 | 29.9 | 35.5 | 52.6 |
| Vision-OPD-9B | 32.6 | 36.3 | 40.0 | 44.2 | 52.7 | 27.8 | 28.2 | 32.5 | 35.9 | 56.4 |
| Open 7–9B models + tool | ||||||||||
| Advantage | 0.5 MP | 1 MP | 2 MP | 4 MP | Max |
|---|---|---|---|---|---|
| ViRA-Bench | |||||
| CrossRPO ( ) | 38.7 (65.1) (12K) | 46.9 (71.1) (16K) | 55.6 (75.8) (26K) | 57.5 (78.2) (44K) | 63.3 (81.3) (147K) |
| Joint term | |||||
| (per-version RLOO) | 31.3 (59.7) (11K) | 41.2 (66.8) (15K) | 50.0 (72.5) (24K) | 52.9 (74.7) (42K) | 58.7 (78.6) (144K) |
| , one version per set | 31.1 (59.3) (10K) | 40.5 (67.3) (15K) | 49.3 (72.5) (23K) | 53.6 (74.7) (41K) | 58.1 (78.0) (147K) |
| 33.4 (61.4) (16K) | 44.2 (68.9) (22K) | 51.3 (73.7) (35K) | 56.1 (76.5) (60K) | 56.6 (78.1) (192K) | |
| Perception | Reasoning | |||||||||||
| Advantage | OCR | RS | DT | MO | AD | All | OCR | DT | MO | AD | All | Overall |
| Maximum resolution | ||||||||||||
| CrossRPO ( ) | 95.2 | 71.3 | 96.0 | 55.8 | 52.9 | 68.8 | 85.0 | 89.0 | 60.7 | 46.0 | 59.9 | 65.3 |
| Joint term | ||||||||||||
| (per-version RLOO) | 95.2 | 69.3 | 96.0 | 53.3 | 50.3 | 67.1 | 84.0 | 92.0 | 62.7 | 44.2 | 59.6 | 64.1 |
| , one version per set | 96.4 | 68.7 | 93.0 | 57.4 | 53.1 | 68.9 | 88.0 | 89.0 | 56.7 | 46.0 | 59.5 | 65.2 |
| Accuracy | ||||||
| Advantage at 1 MP | O | A | B | Zoomed | On obj. | Zooms |
| CrossRPO ( ) | 77.4 | 68.6 | 67.2 | 99.9 | 86.6 | 1.07 |
| Joint term | ||||||
| (per-version RLOO) | 75.0 | 63.4 | 61.9 | 100.0 | 88.3 | 1.06 |
| , one version per set | 78.2 | 61.9 | 61.6 | 100.0 | 90.0 | 1.02 |
| 77.6 | 66.8 | 62.4 | 97.7 | 82.1 | 1.61 | |
| Variant | Baseline | Factor | Scale |
|---|---|---|---|
| CrossRPO ( ) | version | batch | |
| (per-version RLOO) | version | batch | |
| , one version per set | version | batch | |
| version | batch | ||
| version | batch | ||
| Random partners | version | batch |
| V ∗ | HR-4K | HR-8K | MME-RW | TreeBench | O3-Bench | MMStar | Hallusion | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Advantage | 1 | Max | 1 | Max | 1 | Max | 1 | Max | 1 | Max | 1 | Max | Max | Max |
| CrossRPO ( ) | 92.0 | 96.2 | 87.5 | 91.0 | 84.7 | 90.1 | 63.6 | 65.3 | 56.3 | 57.0 | 50.4 | 67.0 | 77.1 | 79.6 |
| Per-version GRPO, | 88.7 | 95.8 | 83.9 | 87.8 | 80.8 | 88.6 | 56.1 | 59.2 | 54.6 | 53.1 | 43.5 | 67.8 | 78.3 | 82.9 |
| Per-version GRPO, | 87.3 | 94.2 | 85.7 | 91.7 | 83.2 | 89.8 | 63.1 | 64.5 | 51.4 | 50.1 | 35.9 | 53.9 | 75.0 | 77.1 |
| Per-version GRPO | 89.5 | 95.3 | 86.0 | 90.8 | 84.6 | 88.8 | 59.5 | 60.6 | 53.6 | 54.8 | 43.8 | 58.6 | 75.1 | 79.2 |
| V ∗ | HR-4K | HR-8K | MME-RW | TreeBench | O3-Bench | MMStar | Hallusion | |||||||
| Run | 1 | Max | 1 | Max | 1 | Max | 1 | Max | 1 | Max | 1 | Max | Max | Max |
| CrossRPO ( ) | 92.0 | 96.2 | 87.5 | 91.0 | 84.7 | 90.1 | 63.6 | 65.3 | 56.3 | 57.0 | 50.4 | 67.0 | 77.1 | 79.6 |
| (per-version RLOO) | 93.0 | 95.6 | 87.8 | 92.9 | 85.7 | 89.8 | 62.4 | 64.1 | 54.3 | 55.3 | 43.5 | 64.1 | 78.8 | 80.2 |
| , one version per set | 90.9 | 95.5 | 86.0 | 92.2 | 82.5 | 90.2 | 63.3 | 65.2 | 56.3 | 56.5 | 43.8 | 63.2 | 78.7 | 78.7 |
| Accuracy | ||||||
| Model at 1 MP | O | A | B | Zoomed | On obj. | Zooms |
| Open 7–9B models + tool | ||||||
| Qwen3.5-9B + tool | 63.4 | 52.5 | 49.9 | 99.1 | 62.6 | 1.82 |
| ZwZ2-9B + tool | 68.9 | 60.5 | 55.8 | 97.1 | 64.8 | 1.56 |
| Vision-OPD-9B + tool | 42.7 | 35.3 | 29.3 | 4.3 | 3.4 | 0.08 |
| Qwen2.5-VL-7B + DeepEyes tool | 47.0 | 24.9 | 19.0 | 49.0 | 18.8 | 2.45 |
| ViRA-9B | Qwen3.5-9B + tool | |||
| 1 MP | Max | 1 MP | Max | |
| Wrong answers (% of images) | 28.9 | 18.7 | 44.8 | 37.1 |
| Share of the wrong answers (%) | ||||
| No answer | 0.7 | 0.5 | 16.6 | 6.4 |
| Answered without zooming | 0.4 | 0.1 | 1.5 | 3.4 |
| No crop covers the object | 21.0 | 10.0 | 22.8 | 19.1 |
| Perception | Reasoning | |||||||||||
| Model at Max | OCR | RS | DT | MO | AD | All | OCR | DT | MO | AD | All | Overall |
| Open 7–9B models | ||||||||||||
| Qwen3.5-9B | 91.2 | 58.7 | 94.0 | 40.8 | 44.0 | 59.4 | 76.0 | 79.0 | 42.7 | 38.2 | 49.6 | 55.5 |
| ZwZ2-9B | 92.0 | 60.7 | 90.0 | 48.0 | 55.4 | 64.8 | 86.0 | 90.0 | 46.0 | 43.2 | 55.7 | 61.3 |
| Vision-OPD-9B | 94.0 | 60.7 | 96.0 | 51.7 | 53.7 | 66.3 | 82.0 | 75.0 | 60.7 | 39.2 | 54.0 | 61.5 |
| Open 7–9B models + tool | ||||||||||||
| Perception | Reasoning | |||||||||||
| Model at 1 MP | OCR | RS | DT | MO | AD | All | OCR | DT | MO | AD | All | Overall |
| Open 7–9B models | ||||||||||||
| Qwen3.5-9B | 85.2 | 50.7 | 83.0 | 39.8 | 43.4 | 55.7 | 76.0 | 72.0 | 42.7 | 37.5 | 48.3 | 52.8 |
| ZwZ2-9B | 86.4 | 55.3 | 82.0 | 44.5 | 50.3 | 59.8 | 82.0 | 76.0 | 46.0 | 42.2 | 52.8 | 57.1 |
| Vision-OPD-9B | 88.0 | 56.7 | 84.0 | 48.0 | 53.4 | 62.4 | 79.0 | 66.0 | 50.7 | 38.8 | 50.1 | 57.6 |
| Open 7–9B models + tool | ||||||||||||
Visual Question Answering
Visual Question Answering (VQA) develops systems that answer natural-language questions about images, videos, and visual documents.
See all 272 papers