cs.CVOct 15, 2026
SaveZoomCap: Evidence-Routed Visual Zooming for Recoverable Visual Information Transfer
Organizations: Ant Group · Zhejiang University
Abstract
Detailed captions transfer visual information into vision-language model supervision, but quality is easily conflated with length: long single-pass outputs still miss small objects, text, and spatial relations or promote uncertain observations into confident claims. We formulate fine-grained captioning as recoverable visual information transfer, in which a teacher acquires missing evidence under a finite observation budget and converts it into reusable supervision. ZoomCap uses a draft caption to route conditional knowledge retrieval, OCR-sensitive reading, coverage-gap recovery, and adaptive visual zooming, allocating deeper crops only to the most complex eligible region, deriving spatial relations from grounded geometry, and fusing non-duplicate evidence. We also introduce FactCycle, which scores caption-derived atomic claims on two blind anchors--source image (truthfulness) and reconstruction (realizability)--separating fabrication from generator loss claim by claim; human and multi-judge comparisons complement this fact-level view. Against Qwen3.5-397B-A17B captions, ZoomCap achieves an 81.9% human tie-adjusted preference score with a teacher over 10x smaller, and FactCycle counts 52.4 transferred claims per image vs. 31.5 for ScaleCap. The evidence also transfers: in continual pre-training, ZoomCap-450K improves the 11-benchmark average by 1.31-2.38 points over the strongest competing corpus and Ling-Tiny-VL-v3 gains 1.4 points on a production document/OCR monitor suite (1.5 on perception) at 60B tokens; in medical pre-training, 2.64M ZoomCap-captioned images improve five VQA benchmarks by 3.04 points (vs. 1.76 for same-image Gemini-3.1-Pro captions).
Figures & tables
Figure 1: Evidence preservation comparisons. (a) On the same image, five targets remain partial or missing from ScaleCap ( Xing et al., 2025 ) and Gemini-3.1-Pro ( Google DeepMind, 2026 ) but are retained by ZoomCap. (b) ZoomCap-450K has 1.45 ScaleCap’s total characters and a 1.58 larger P99 character length.
Figure 2: ZoomCap overview. From a raw image and a draft caption, ZoomCap routes each sample through agentic evidence acquisition: external-knowledge search, OCR extraction for text-, chart-, and document-like images, and adaptive visual zoom-in for detail-rich scenes. The zoom-in branch combines grounded boxes, semantic regions, caption-coverage gaps, recursive patch observations, and A-SPOE relation facts before multi-source fusion produces the final caption.
Figure 3: Adaptive zoom-in mechanisms. Left: rule complexity nominates dense parents while the agent rejects uninformative zooms; only the highest-scoring parent approved by both expands. Center: caption-coverage gaps derived from draft-grounded entity boxes trigger targeted re-observation (example from PrefBench-456). Right: A-SPOE retains only relations supported by the geometric margin.
Figure 4: FactCycle under its two blind anchors. The same blind-extracted claim set is scored against the source image (Truthfulness ) and against the fixed-generator reconstruction (Realizability ); joining the two verdicts per claim yields four cells, and the headline aggregates the transferred cell.
Figure 5: Information density and generative recoverability in FactCycle. (a) each point is one FactCycle-250 image with per-source linear fits; ZoomCap occupies the high-support, high-similarity corner. (b) mean extracted claims per caption: 67.1 vs. 41.9 and 20.7.
| Caption source | #Facts/img | : Truthfulness (%) | transferred (%) | specious (%) | unrendered fab. (%) | DINO | |
| ShareGPT4V-450K ( Chen et al., 2024a ) | 20.7 | 81.4 | 80.3 | 16.6 | 2.0 | 14.8 | 0.566 |
| ScaleCap-450K ( Xing et al., 2025 ) | 41.9 | 88.9 | 88.2 | 7.4 | 2.6 | 31.5 | 0.620 |
| ZoomCap-450K (Ours) | 67.1 | 93.8 | 90.9 | 3.4 | 1.7 | 52.4 | 0.696 |
Table 1: FactCycle master table on FactCycle-250 (two anchors, one blind claim set). The same blind claim set is scored on both anchors by the same blind judge pair ( Kimi-K3 , Qwen3.8-Max ) under one supported/unsupported/unverifiable rubric; neither judge sees the caption, its corpus identity, or the other anchor’s image (protocol in Appendix E ). #Facts/img counts blind-extracted claims; and follow equation 6 –equation 8 ; the transferred/specious/unrendered-fabrication shares, with the excluded true-but-unrendered generator cell, partition the claims decidable on both anchors. is prose-only (Section 4.1.1 ). DINO is DINOv3 ( Siméoni et al., 2025 ) source–reconstruction similarity, a judge-independent global check.
Figure 6: Paired-preference evaluation against Qwen3.5-397B-A17B on PrefBench-456. ZoomCap’s largest gains are in knowledge richness and information coverage.
| Configuration | Final score | Visual accuracy | Information coverage | Knowledge richness | Avg. local crops |
| (a) Evidence streams — opponent: Qwen3.5-397B-A17B single-pass caption | |||||
| Global draft only (no routed evidence) | 55.9 | 57.9 | 41.6 | 47.6 | 0.0 |
| Agentic evidence w/o retrieved knowledge | 53.0 | 53.0 | 62.4 | 46.7 | 5.0 |
| Agentic evidence w/ retrieved knowledge | 74.2 | 58.0 | 69.8 | 92.3 | 5.0 |
| (b) Zoom budget — direct head-to-head; the two rows are its two sides | |||||
| Full second-level pyramid | 28.8 | 36.3 | 23.4 | 50.1 | 10.13 |
Table 2: Evidence-stream attribution (a) and zoom-budget policy (b) ablations on PrefBench-456; cells are tie-adjusted preference scores (%) (wins plus half ties; sides sum to 100). (a) judges each configuration against the same Qwen3.5-397B-A17B single-pass caption (first-level crops only); (b) fixes the streams and pairs the two zoom policies directly (rows = two sides of one head-to-head; crops include the second level).
| Doc & OCR suite | Perception suite | ||||
| Bench. | Base | + ZoomCap | Bench. | Base | + ZoomCap |
| ChartQA ( Masry et al., 2022 ) | 83.7 | 83.9 (+0.2) | AI2D ( Kembavi et al., 2016 ) | 82.1 | 81.8 (-0.2) |
| DocVQA ( Mathew et al., 2021 ) | 95.4 | 95.6 (+0.1) | eyes_general | 0.86 | 0.83 (-0.03) |
| InfoVQA ( Mathew et al., 2022 ) | 80.2 | 80.4 (+0.2) | eyes_vertical | 71.1 | 71.2 (+0.1) |
| OCRBench ( Liu et al., 2024c ) | 86.4 | 86.7 (+0.3) | hallusion_intern ( Guan et al., 2024 ) | 58.6 | 60.0 (+1.4) |
| OCRBenchV2-a ( Fu et al., 2025 ) | 40.1 | 40.1 (+0.1) | MMBench ( Liu et al., 2024b ) | 78.6 | 79.1 (+0.5) |
Table 3: General-domain production adoption (Ling-Tiny-VL-v3). Matched 60B-token runs share checkpoint, hyperparameters, and seed; the experimental arm swaps the production vl_caption snapshot ( pretrain_caption_20260623 ) for zoomcap_pretrain_1600w (16M ZoomCap captions, up-sampled to a 20% caption-group share); sub-items average the four common 40–55B checkpoints.
| Medical pretraining recipe | Medical images | Hallusion (intern) | GMAI MMBench | PathVQA | PMC VQA | SLAKE | VQA RAD | Avg. gain |
| Reference recipe (128 H800) | 3.66M | 60.9 | 55.3 | 39.4 | 60.9 | 68.6 | 64.3 | – |
| + Gemini-3.1-Pro captions (64 H800) | +2.64M | 59.4 | 59.3 | 40.0 | 62.8 | 71.3 | 63.9 | +1.76 |
| + ZoomCap captions (Ours; 64 H800) | +2.64M | 60.4 | 61.4 | 39.7 | 64.2 | 71.0 | 67.4 | +3.04 |
Table 4: Medical-domain production adoption (Ling-Tiny-VL-v3). Gemini-3.1-Pro and ZoomCap caption the same 2.64M images under one 64-H800, 50B-token recipe (the 128-H800 reference is not compute-matched); Avg. gain covers the five medical VQAs.
| Model | Pretraining Data | Info VQA | Doc VQA | Chart QA | MM Star | Math Vista | LLaVA Bench | MMVet | MMB | MMMU | SEED | AI2D | Avg. |
| Qwen2.5-7B + Qwen2.5-ViT | Vanilla | 43.66 | 77.25 | 73.56 | 45.93 | 47.20 | 51.70 | 48.35 | 72.47 | 44.78 | 71.80 | 74.03 | 59.16 |
| ShareGPT4V-450K | 45.80 | 79.46 | 74.20 | 48.87 | 49.60 | 44.70 | 49.22 | 72.52 | 42.89 | 72.38 | 74.38 | 59.46 | |
| DenseFusion-450K | 47.94 | 81.48 | 74.36 | 50.80 | 45.20 | 54.00 | 48.17 | 72.05 | 45.00 | 71.66 | 74.16 | 60.44 | |
| ScaleCap-450K | 49.02 | 81.94 | 74.96 | 50.39 | 46.60 | 50.00 | 47.39 | 73.71 | 43.29 | 72.32 | 75.68 | 60.48 | |
| ZoomCap-450K (Ours) | 51.08 | 82.64 | 74.76 | 52.66 | 49.30 | 54.80 | 52.75 | 74.65 | 48.40 | 72.72 | 75.94 | 62.70 | |
| Qwen2.5-3B + Qwen2.5-ViT | Vanilla | 37.35 | 70.11 | 70.24 | 47.13 | 44.70 | 43.80 | 42.20 | 68.36 | 39.00 | 69.56 | 70.76 | 54.84 |
Table 5: CPT comparison with different 450K caption datasets on 11 benchmarks. Within each backbone block only the CPT caption corpus varies; all methods are re-evaluated with Qwen3.5-397B-A17B for judge-dependent benchmarks.
Figure 7: Qualitative FactCycle reconstruction comparison. Each row uses captions of the same source image to condition the fixed text-to-image generator; ZoomCap reconstructions preserve more of the source composition and fine-grained content; the remaining fourteen FactCycle-250 pairs appear in Appendix Figure 9 .
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
| Scene | Caption source | #Facts/img | (%) | transferred (%) | specious (%) | unrendered fab. (%) | |
| Indoor | ShareGPT4V-450K ( Chen et al., 2024a ) | 19.6 | 81.6 | 79.0 | 17.3 | 2.4 | 14.3 |
| ScaleCap-450K ( Xing et al., 2025 ) | 40.2 | 88.6 | 87.1 | 8.3 | 2.6 | 30.3 | |
| ZoomCap-450K (Ours) | 68.8 | 93.2 | 88.6 | 4.0 | 2.0 | 52.4 | |
| Outdoor nature | ShareGPT4V-450K ( Chen et al., 2024a ) | 19.7 | 81.2 | 78.7 | 18.7 | 1.6 | 14.0 |
| ScaleCap-450K ( Xing et al., 2025 ) | 39.0 | 89.5 | 89.5 | 6.3 | 2.5 | 30.1 | |
| ZoomCap-450K (Ours) | 58.8 | 94.1 | 91.8 | 3.6 | 1.6 | 46.3 |
Table 7: Two-anchor FactCycle scoring per scene bucket (50 images each). Columns follow Table 1 : #Facts/img counts blind-extracted claims, Truthfulness is supported/(supported+unsupported) against the source image; the transferred, specious, and unrendered-fabrication shares cover claims decidable on both anchors and, with the excluded true-but-unrendered generator cell, sum to 100%. Per-bucket realizability is discussed in the Realizability accounting paragraph below.
| Caption source | Metric | entity | attribute | spatial | count | small obj. | action | other |
| ShareGPT4V-450K | transferred (%) | 89.5 | 78.8 | 77.4 | 68.0 | 78.7 | 70.7 | 86.7 |
| specious (%) | 10.1 | 18.5 | 18.5 | 20.1 | 19.8 | 24.1 | 8.6 | |
| unrendered fab. (%) | 0.1 | 1.5 | 3.1 | 6.4 | 0.9 | 3.3 | 3.6 | |
| ScaleCap-450K | transferred (%) | 94.2 | 87.8 | 82.8 | 83.8 | 85.4 | 84.8 | 93.6 |
| specious (%) | 4.8 | 8.2 | 9.4 | 7.0 | 8.4 | 8.1 | 4.0 | |
| unrendered fab. (%) | 0.5 | 2.4 | 4.5 | 4.9 | 3.7 | 4.2 | 1.8 |
Table 8: Two-anchor FactCycle quadrant shares by fact category. Shares are computed over claims decidable on both anchors within each category: transferred (true and rendered), specious (fabricated yet rendered), and unrendered fabrication (fabricated and unrendered); the true-but-unrendered generator cell is excluded from caption metrics. Shading marks the best cell of each column (highest transferred, lowest specious and lowest unrendered fabrication).
| Caption source | Fabricated-then- rendered (%) | True-among- rendered (%) | Fabrication rate (%) |
| ShareGPT4V-450K ( Chen et al., 2024a ) | 88.1 | 79.7 | 18.4 |
| ScaleCap-450K ( Xing et al., 2025 ) | 72.5 | 87.0 | 9.9 |
| ZoomCap-450K (Ours) | 64.7 | 92.9 | 5.1 |
Table 9: Conservative dual-anchor accounting on FactCycle-250. In the conservative variant, reconstruction-leg unverifiable verdicts count toward the unrealized denominator. Fabricated-then-rendered: share of claims unsupported by the source image that the reconstruction nevertheless renders; true-among-rendered: share of rendered claims that the source image also supports; fabrication rate: share of claims unsupported by the source image under the conservative denominator. Lower is better for all three.
Figure 10: Checkpoint-level curves for the general-domain adoption study. Suite means over the shared checkpoints (40–55B tokens) for the baseline mixture and the mixture with ZoomCap captions: (a) the nine-benchmark Doc & OCR suite, (b) OmniDocBench, (c) the nine-benchmark perception suite; the gains are stable across checkpoints.
More in this topic
Image Captioning
Image captioning generates natural-language descriptions of visual content, with research focused on accuracy, coverage, grounding, and evaluation.
See all 96 papers