{"id":"8424cda8-8bd2-4bc7-a62d-d4a99cd7b240","arxiv_id":"2508.09650","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"A temporal, occlusion-aware ball-tracking network reduces RMSE from 37.30 to 7.19 and improves fully-occluded detection accuracy from 0.63 to 0.80 across four sports datasets.","lead":"TOTNet is a neural network for detecting and tracking a small ball in sports videos when it is hidden or occluded. It cuts tracking error sharply on table tennis, tennis, and badminton videos and ships with a new occlusion-rich table tennis dataset.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Headline performance gain depends on unverified protocol comparability; without identical splits/occlusion definitions, the RMSE reduction is not interpretable.","rationale":"The reader's UNVERDICTED verdict is appropriate because the full text is inaccessible and the abstract alone cannot validate the comparison. My stress-test identifies the same load-bearing assumption: the reported performance gain is only meaningful if all methods were compared under identical conditions. The concrete test I propose would settle this by reproducing the baseline numbers under TOTNet's evaluation protocol. I do not allege any misconduct; the issue is that the current evidence does not yet rule out protocol differences as the source of the improvement. Therefore the verdict remains UNVERDICTED pending full-text and reproducibility review.","tokens_in":1790,"tokens_out":1654,"duration_ms":22845,"concrete_test":"Using the public repository (AugustRushG/TOTNet), rerun the full evaluation protocol: for each of the four datasets, evaluate every prior SOTA method with its official code/checkpoints on the exact TOTNet test splits and the exact occlusion masks used to compute the reported metrics. If the baseline RMSE of 37.30 and fully-occluded accuracy of 0.63 are reproduced within a small tolerance (e.g., ±5%) under this uniform protocol, the central claim is supported. If baselines' numbers shift materially, the claimed margin is a protocol artifact rather than an algorithmic advantage.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract's central claim is that TOTNet reduces RMSE from 37.30 to 7.19 and raises fully-occluded-frame accuracy from 0.63 to 0.80 across four datasets. For this to demonstrate algorithmic superiority, every prior SOTA method must be evaluated under identical train/test splits, identical occlusion annotations, and identical post-processing. The supplied full text is garbled, so the methods section and experimental protocol cannot be inspected. The reported numbers therefore rest on an unverified premise: that the comparison is apples-to-apples. A single protocol difference—for example, baselines evaluated without the same occlusion-augmentation training, or using a different definition of 'fully occluded'—could account for a large fraction of the reported gap. This is a correctness risk, not a novelty or consensus dispute: the claim may be true, but the evidence as currently available is insufficient to establish it.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes TOTNet, a temporal occlusion tracking network for ball tracking in sports videos, combining 3D convolutions, visibility-weighted loss, and occlusion augmentation. It also introduces TTA, an occlusion-rich table tennis dataset of 9,159 samples with 1,996 occlusion cases collected from Paralympic matches. The abstract reports large performance improvements over prior state-of-the-art, reducing RMSE from 37.30 to 7.19 and increasing fully-occluded-frame accuracy from 0.63 to 0.80 across four datasets in tennis, badminton, and table tennis. The supplied full text, however, is largely illegible (character-encoding corruption), so the architecture details, equations, dataset splits, and experimental protocol cannot be inspected. This report therefore assesses the abstract-level claims and the completeness of the evidence.","tokens_in":2043,"tokens_out":3801,"duration_ms":37954,"significance":"If the reported results are reproducible, TOTNet would be a practically valuable contribution to sports-video ball tracking under occlusion, and the TTA dataset plus the announced code/data release would help the community. The collaboration with Paralympics Australia gives the work a real-world grounding. However, the significance is conditional: the headline numbers are not interpretable without a verifiable evaluation protocol. As submitted, the evidence for the central claim is incomplete because the baselines, splits, occlusion definitions, and post-processing are not available for inspection.","major_comments":[{"comment":"The central claim rests on comparing TOTNet with prior state-of-the-art methods, yet no protocol is given: train/test splits, evaluation metrics, baseline versions, hyperparameters, or post-processing. Without evidence that all methods were evaluated under identical conditions, the reported RMSE drop from 37.30 to 7.19 and the fully-occluded accuracy rise from 0.63 to 0.80 cannot be attributed to the proposed method. This is load-bearing; please provide a full experimental setup, per-dataset results, and baseline configurations.","section":"Abstract (comparison protocol)"},{"comment":"The paper introduces TTA with 9,159 samples and 1,996 occlusion cases but does not state how samples are split, how occlusion is defined or annotated, or how TTA is used relative to the other three datasets. If TTA is used for training or evaluation, a leakage/overlap check and class distribution should be reported. Without this, the dataset's contribution and the comparability of the reported fully-occluded accuracy cannot be assessed.","section":"Abstract (TTA dataset and evaluation)"},{"comment":"The body of the submitted manuscript is largely unreadable (garbled character encoding), so the 3D convolution design, the visibility-weighted loss, the occlusion-augmentation procedure, and the training details cannot be inspected. This is not a merely stylistic issue: the central derivation and the implementation needed to reproduce the claims are inaccessible. A clean, fully rendered version of the manuscript is required.","section":"Full text / Methods"},{"comment":"No error bars, multiple-seed results, or statistical significance tests are reported. For a performance improvement of the reported magnitude, the paper should show variance across runs and ablations. Without ablations isolating the 3D convolution, the visibility-weighted loss, and the occlusion augmentation, it is unknown which component drives the gain or whether the gain is stable.","section":"Evaluation / statistical validity"}],"minor_comments":[{"comment":"Typo: 'TOTNets effectiveness' should be 'TOTNet's effectiveness'.","section":"Abstract"},{"comment":"The code/data statement mentions a GitHub link but no license; please specify the license and what exactly is released (dataset annotations, pretrained weights, evaluation scripts).","section":"Abstract / Code release"},{"comment":"The abstract says the method is for 'offline sports analytics' but also emphasizes 'fast-paced scenarios'; please clarify the runtime assumptions and whether the method is intended for offline use only.","section":"Abstract / Scope"}],"recommendation":"major_revision","confidential_remarks":"The submitted full text is illegible, which alone prevents a proper technical review. I recommend major revision: the authors must supply a readable manuscript with a complete methods section, detailed evaluation protocol, per-dataset results, ablations, and error bars. If the illegibility is a rendering artifact, the arXiv PDF should be regenerated before resubmission."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nQuick take on TOTNet: the architecture and dataset are real contributions, but the abstract's numbers are not something I'd rely on before seeing the experimental protocol. The RMSE drop from 37.30 to 7.19 is a 5x improvement, and fully-occluded accuracy jumping from 0.63 to 0.80 is similarly large. Those figures are either the result of a genuinely better method or a mismatch in evaluation. The abstract gives no way to tell.\n\nWhat's genuinely new is the TTA dataset: 9,159 labeled samples from professional Paralympic table tennis, with 1,996 occlusion cases. That's a solid resource for a domain where occlusion is common. The method itself—3D temporal convolutions, a visibility-weighted loss, and occlusion augmentation—is a sensible combination for occlusion-aware tracking. The collaboration with Paralympics Australia also gives it a concrete real-world anchor.\n\nThe soft spot is the evaluation. We only have the abstract, and it doesn't specify train/test splits, baseline configurations, post-processing, or how 'fully occluded' is defined. Without those, the headline comparison is unverifiable. The stress-test note is exactly on target: if prior methods weren't trained under the same occlusion augmentation, or if occlusion definitions differ, a large chunk of the reported gap could evaporate. That's not an indictment—the method could still be better—but it's a load-bearing unknown. I'd also like to see error bars or multiple-run averages.\n\nWho is this for? Researchers working on ball tracking in sports video, especially under occlusion. The TTA dataset alone may be worth the paper even if the architecture's gains don't fully hold up.\n\nMy recommendation: send it to peer review. A serious referee should demand the full protocol, check baselines and occlusion definitions, and verify against the promised code and data. This is a paper worth engaging with, not a desk reject.","headline":"Real contributions in the dataset and architecture, but the headline accuracy gains are uninterpretable without the full evaluation protocol.","tokens_in":2466,"tokens_out":2194,"would_cite":false,"duration_ms":23328,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"TOTNet tracks balls through full occlusion, cutting positional RMSE from 37.30 to 7.19 across four sports datasets.","keywords":["ball tracking","occlusion-aware tracking","3D convolutions","sports video analysis","visibility-weighted loss","temporal modeling","table tennis dataset","badminton"],"falsifier":"Run the earlier trackers on the same train/test splits and the same occlusion label definitions as TOTNet, with identical post-processing; if their RMSE on fully occluded frames approaches 7.19 or the accuracy gap toward 0.80 closes, the claimed superiority is protocol-driven rather than algorithmic.","tokens_in":1761,"feed_emoji":"🎾","tokens_out":5144,"duration_ms":46963,"temperature":0.7,"pith_summary":"The paper sets out to solve ball tracking in sports video when the ball is partially or fully hidden. It proposes TOTNet, a network that uses 3D convolutions over short clips, a visibility-weighted loss, and occlusion augmentation during training. Across four datasets spanning tennis, badminton, and table tennis, the authors report reducing tracking RMSE from 37.30 to 7.19 and raising accuracy on fully occluded frames from 0.63 to 0.80 compared with earlier methods. If these numbers hold, offline sports analytics gains a way to keep following the ball through the moments when referees and event systems most need it.","feed_headline":"Occlusion-aware tracker cuts ball-position error from 37.30 to 7.19","feed_subtitle":"Full-frame occlusion accuracy jumps from 0.63 to 0.80 across tennis, badminton, and table tennis datasets.","key_machinery":"The load-bearing mechanism is a 3D convolutional network operating over a short temporal window of frames, trained with a visibility-weighted loss and occlusion augmentation. The visibility-weighted loss up-weights frames with known ball positions so uncertain or missing labels do not dominate the gradient; occlusion augmentation randomly hides the ball during training so the model must predict trajectories through invisible spans. The 3D convolutions are what let the network exploit motion and trajectory priors across the window.","core_discovery":"The paper's central claim is that occlusion-resilient ball tracking is best achieved by giving the model explicit temporal context rather than a stronger per-frame detector. TOTNet treats a short window of frames as a 3D volume, weights the loss by whether the ball is visible, and trains on clips where occlusion is artificially inserted. The reported result is a large drop in tracking RMSE—from 37.30 to 7.19—and a jump in fully occluded frame accuracy from 0.63 to 0.80 on four datasets in tennis, badminton, and table tennis. The paper also introduces TTA, a new occlusion-rich table tennis dataset with 9,159 samples, 1,996 of them occlusion cases, drawn from professional-level matches.","pith_inferences":["Editorial extension: the same temporal occlusion-handling mechanism should generalize to other small, fast objects such as shuttlecocks, cricket balls, or pucks, provided training clips are long enough for the network to learn each object's bounce dynamics.","Editorial extension: if the reported RMSE drop holds, single-frame appearance-based detectors are probably at a structural disadvantage under occlusion; future gains are more likely to come from trajectory priors and smoothing than from stronger per-frame features.","Editorial extension: the visibility-weighted loss could be repurposed for semi-supervised tracking, where only a subset of frames in each clip carries ball labels, potentially reducing annotation cost."],"forward_implications":["Offline ball tracking in tennis, badminton, and table tennis can preserve ball position through full occlusion, supporting more reliable automatic line calls and event detection.","The visibility-weighted loss offers a way to train trackers when ground-truth ball positions are available for only some frames in a clip.","Occlusion augmentation can be applied to existing tracking datasets without new data collection, making the approach transferable to other sports with labeled footage.","The new table tennis dataset with 9,159 samples, including 1,996 occlusion cases, provides a benchmark focused specifically on occlusion-heavy ball tracking."],"supporting_citations":[],"fun_headline_variants":["Ball tracking error slashed 80% with temporal occlusion awareness","Occlusion-aware 3D tracker cuts RMSE from 37.3 to 7.2","Fully occluded ball detection accuracy jumps to 0.80","TOTNet beats state-of-the-art on occlusion-heavy sports video","New table tennis dataset drives robust ball tracking gains"],"cache_read_input_tokens":2816,"weakest_assumption_plain":"The comparisons assume earlier methods were evaluated on the same train/test splits, the same occlusion definitions, and the same post-processing, so the reported RMSE drop and accuracy gain reflect the algorithm rather than evaluation-protocol differences.","fun_headline_variants_meta":{"raw":{"variants":["Ball tracking error slashed 80% with temporal occlusion awareness","Occlusion-aware 3D tracker cuts RMSE from 37.3 to 7.2","Fully occluded ball detection accuracy jumps to 0.80","TOTNet beats state-of-the-art on occlusion-heavy sports video","New table tennis dataset drives robust ball tracking gains"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000896,"raw_usage":{"total_tokens":3703,"prompt_tokens":752,"completion_tokens":2951,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":496,"completion_tokens_details":{"reasoning_tokens":2867}},"tokens_in":496,"tokens_out":2951,"duration_ms":21981,"temperature":1.0,"reasoning_tokens":2867,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T20:53:31.904115+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the earlier trackers on the same train/test splits and the same occlusion label definitions as TOTNet, with identical post-processing; if their RMSE on fully occluded frames approaches 7.19 or the accuracy gap toward 0.80 closes, the claimed superiority is protocol-driven rather than algorithmic.","supporting_citations":[],"review_version":1}