{"id":"d779d525-7070-4508-bd31-ec891e1d206d","arxiv_id":"2606.08063","paper_version":1,"verdict":"CONDITIONAL","confidence":"LOW","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Robust-U1 equips MLLMs with self-recovery via supervised fine-tuning, RL using SSIM and CLIP rewards, and joint multimodal reasoning, yielding SOTA robustness on corruption benchmarks.","lead":"The paper presents Robust-U1, a framework that trains multimodal LLMs to reconstruct corrupted images themselves using fine-tuning followed by reinforcement learning with pixel and semantic rewards. Smart readers might care because it offers a self-contained way to make vision-language models more reliable under real-world image damage without needing separate restoration tools.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Causal contribution of RL-recovered images to reasoning gains remains unisolated","rationale":"The reader's weakest assumption directly identifies the missing causal isolation between RL recovery quality and downstream reasoning; the proposed ablation would test exactly that link without requiring new training.","tokens_in":1764,"tokens_out":288,"duration_ms":8754,"concrete_test":"In the multimodal reasoning stage, replace the RL-recovered image with (a) the original corrupted input or (b) a random noise image of matching size while keeping all other inputs and prompts identical; recompute VQA accuracy on the real-world corruption benchmark. If the drop is <2 points relative to the reported RL-recovered setting, the self-recovery mechanism is not load-bearing.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The strongest claim requires that the dual-reward RL stage (SSIM + CLIP) produces recoveries whose pixel/semantic content causally drives the reported robustness gains in the final multimodal reasoning stage. This is least secure because the reasoning prompt jointly feeds both corrupted and recovered images; any performance lift could arise from the joint format, from the SFT stage, or from metric overfitting rather than genuine visual restoration. No explicit ablation (e.g., substituting the RL output with the original corruption, a constant image, or an SFT-only output) is described that would falsify this alternative.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper proposes Robust-U1, a three-stage framework equipping MLLMs with self-recovery capability for corrupted images: supervised fine-tuning for initial reconstruction, reinforcement learning using dual rewards (pixel-level SSIM and semantic-level CLIP similarity), and multimodal reasoning that jointly processes the corrupted input and recovered image. It claims this yields state-of-the-art robustness on real-world corruption benchmarks and superior performance under adversarial corruptions on VQA tasks, with analysis showing that high-quality visual recovery directly enhances reasoning.","tokens_in":1858,"tokens_out":437,"duration_ms":13129,"significance":"If validated, the result would establish an interpretable pixel- and semantic-level self-recovery mechanism as a critical component for MLLM robustness, going beyond black-box feature alignment or text-only reasoning; the public source code link supports reproducibility.","major_comments":[{"comment":"The multimodal reasoning stage jointly feeds both corrupted and recovered images, yet no ablation is described (e.g., substituting RL output with original corruption, a constant image, or SFT-only output) to isolate whether reported reasoning gains are causally driven by the RL-recovered content rather than the joint prompt format or prior stages; this directly affects the central claim that recovery enhances reasoning.","section":"Abstract / multimodal reasoning stage"},{"comment":"No quantitative details are given on the relative weighting between SSIM and CLIP rewards in the RL stage, nor on ablation controls or error bars for the SOTA results; this leaves the internal validity of the experimental outcomes unverifiable from the reported high-level outcomes.","section":"Reinforcement learning stage"}],"minor_comments":[{"comment":"The abstract states that source code is available; ensure the repository includes exact hyperparameter values for the dual-reward weighting and the full set of benchmark numbers with error bars.","section":null}],"recommendation":"major_revision","confidential_remarks":"The provided reader's assessment notes low confidence stemming from missing quantitative controls; this aligns with the need for explicit ablations before the causal claim can be considered secure."},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive feedback and positive evaluation of the work's potential significance. We address each major comment below with plans for targeted revisions to improve clarity and experimental rigor.","responses":[{"response":"We agree that the absence of these ablations limits the strength of the causal claim. In the revised version we will add a dedicated ablation subsection reporting performance when the RL-recovered image is replaced by (i) the original corrupted input, (ii) a constant image, and (iii) the SFT-stage output only, while keeping the joint-prompt format fixed. These results will directly quantify the incremental benefit attributable to the RL-recovered content.","revision_made":"yes","referee_comment":"[Abstract / multimodal reasoning stage] The multimodal reasoning stage jointly feeds both corrupted and recovered images, yet no ablation is described (e.g., substituting RL output with original corruption, a constant image, or SFT-only output) to isolate whether reported reasoning gains are causally driven by the RL-recovered content rather than the joint prompt format or prior stages; this directly affects the central claim that recovery enhances reasoning."},{"response":"We will insert the exact weighting coefficient (λ) used to combine the SSIM and CLIP terms in the composite reward, together with an ablation table that isolates each reward component. For the main SOTA tables we will also report standard deviations computed over three independent training seeds. These additions will be placed in the experimental section and appendix.","revision_made":"yes","referee_comment":"[Reinforcement learning stage] No quantitative details are given on the relative weighting between SSIM and CLIP rewards in the RL stage, nor on ablation controls or error bars for the SOTA results; this leaves the internal validity of the experimental outcomes unverifiable from the reported high-level outcomes."}],"tokens_in":1353,"tokens_out":401,"duration_ms":16208,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main point is that Robust-U1 trains an MLLM to reconstruct corrupted images first via supervised fine-tuning, then refines the output with reinforcement learning that balances SSIM pixel reward and CLIP semantic reward, and finally runs reasoning on both the original corrupted image and the recovered one. This setup produces state-of-the-art results on a real-world corruption benchmark and holds up under adversarial cases on standard VQA tasks.\n\nWhat is new is the specific combination of internal visual self-recovery inside an MLLM using that dual-reward RL stage before joint reasoning. The code is released, which is useful for anyone wanting to reproduce or extend the pipeline.\n\nThe work does a reasonable job demonstrating that an MLLM can be made to produce its own recovered visuals and that doing so lines up with better downstream performance in their reported experiments. The choice of SSIM plus CLIP as rewards is a straightforward way to target both low-level fidelity and higher-level content.\n\nThe soft spot is that the central claim about recovery driving the gains rests on thin evidence for causality. The reasoning stage feeds both the corrupted input and the recovered image together, so lifts could come from the joint prompt format, the supervised stage alone, or even metric overfitting rather than genuine visual restoration. The abstract mentions an analysis confirming the benefit, but no explicit ablations are described that would rule out the alternatives, such as feeding a dummy or SFT-only image instead. Details on reward weighting, controls, and error bars are also missing from the summary, which makes the internal validity harder to judge.\n\nThis is for researchers working on robustness in multimodal models, especially those already using RL or recovery techniques in computer vision. A reader focused on practical fixes for corrupted inputs in VQA would get value from the pipeline description and the public code. The concrete method and code release are enough to merit a serious referee who can check the ablations and run the numbers.\n\nI would send it to peer review.","headline":"The paper adds a three-stage self-recovery pipeline with dual-reward RL to MLLMs and reports SOTA robustness numbers, but the causal role of the recovered images is not isolated.","tokens_in":2410,"tokens_out":483,"would_cite":false,"duration_ms":14084,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"MLLMs can self-recover corrupted visual content to achieve robust understanding.","keywords":["multimodal large language models","visual self-recovery","robustness to corruptions","reinforcement learning","image reconstruction","visual question answering","adversarial robustness"],"falsifier":"An ablation showing that models given only the original corrupted images achieve the same or higher accuracy on the corruption and VQA benchmarks as models given the recovered images.","tokens_in":2622,"feed_emoji":"🖼️","tokens_out":619,"duration_ms":20158,"temperature":0.7,"pith_summary":"The paper asks whether multimodal large language models can restore corrupted images by themselves and shows that doing so improves their reasoning on degraded inputs. It introduces a three-stage process: supervised fine-tuning to reconstruct images, reinforcement learning that rewards both pixel-level similarity and semantic similarity, and joint reasoning that takes both the original corrupted image and the recovered version as input. Experiments report state-of-the-art results on real-world corruption benchmarks and better performance under adversarial corruptions on visual question answering tasks. A sympathetic reader would care because everyday images often contain noise, blur, or other degradations, and the work suggests models can internally fix these issues rather than relying on separate preprocessing or text-only workarounds.","feed_headline":"MLLMs self-recover corrupted images to improve reasoning","feed_subtitle":"Three-stage framework with fine-tuning and dual-reward RL lets models restore details and reason from both damaged and fixed versions.","key_machinery":"Robust-U1 framework with supervised fine-tuning for reconstruction, dual-reward reinforcement learning using SSIM and CLIP, and multimodal reasoning over both corrupted and recovered images.","core_discovery":"Robust-U1 equips MLLMs with explicit visual self-recovery through supervised fine-tuning for initial reconstruction, reinforcement learning with dual rewards of SSIM and CLIP similarity for high visual quality, and multimodal reasoning that jointly considers the corrupted input and the recovered image; experiments confirm that high-quality visual recovery directly enhances reasoning performance.","pith_inferences":["If the recovery step generalizes beyond the tested corruptions, models could handle novel degradations without additional fine-tuning.","Joint use of corrupted and recovered images may let models learn to identify and compensate for specific damage patterns.","The approach could extend to other input modalities such as video frames or audio signals that suffer analogous degradations."],"forward_implications":["State-of-the-art robustness on the real-world corruption benchmark.","Superior performance under adversarial corruptions on general VQA benchmarks.","High-quality visual recovery directly enhances reasoning performance.","Self-recovery functions as a critical mechanism for robust visual understanding."],"fun_headline_variants":["MLLMs self-recover corrupted visuals for robust understanding","MLLMs learn visual self-recovery through fine-tuning and RL","Robust-U1 equips MLLMs with explicit visual self-recovery","MLLMs jointly reason from corrupted and recovered images"],"cache_read_input_tokens":64,"weakest_assumption_plain":"The dual-reward reinforcement learning stage produces image recoveries that genuinely improve downstream reasoning rather than merely optimizing the chosen SSIM and CLIP metrics.","fun_headline_variants_meta":{"raw":{"variants":["MLLMs self-recover corrupted visuals for robust understanding","MLLMs learn visual self-recovery through fine-tuning and RL","Robust-U1 equips MLLMs with explicit visual self-recovery","MLLMs jointly reason from corrupted and recovered images"]},"model":"grok-4.3","cost_usd":0.006272,"raw_usage":{"total_tokens":2946,"prompt_tokens":659,"num_sources_used":0,"completion_tokens":67,"cost_in_usd_ticks":62724500,"prompt_tokens_details":{"text_tokens":659,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2220,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":659,"tokens_out":67,"duration_ms":11889,"temperature":1.0,"reasoning_tokens":2220,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-27T20:04:37.520144+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"An ablation showing that models given only the original corrupted images achieve the same or higher accuracy on the corruption and VQA benchmarks as models given the recovered images.","supporting_citations":[],"review_version":1}