{"id":"6c5f6847-7013-41cf-a45f-5af5ae3c46da","arxiv_id":"2606.23152","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Presents ShotcreteDepth, a dataset of 11,252 synchronized stereo RGB and LiDAR samples from construction sites with 220 annotations for depth tasks in harsh conditions.","lead":"This paper introduces ShotcreteDepth, a new bi-modal dataset of stereo RGB images and LiDAR point clouds collected from active shotcrete construction sites under harsh conditions like poor lighting and high turbidity. Smart generalists and roboticists might read it to see how specialized real-world data can help train perception systems for challenging industrial environments.","discovery_kind":"new_application","skeptic_critique":{"model":"grok-4.3","headline":"Dataset synchronization and harsh condition representation lack quantitative validation","rationale":"The reader's weakest_assumption directly identifies the load-bearing empirical claim for a dataset-introduction paper. Full text adds no quantitative validation of synchronization or condition metrics, so the conservative UNVERDICTED stance remains appropriate.","tokens_in":1715,"tokens_out":251,"duration_ms":15121,"concrete_test":"Download the released dataset and compute timestamp deltas between paired RGB and LiDAR files for 100 random samples; if mean offset exceeds 10 ms or any sample exceeds 30 ms, the synchronization claim is not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that the 11,252 stereo RGB + LiDAR samples are temporally synchronized and faithfully capture high turbidity/poor illumination. The manuscript states acquisition occurred under these conditions and that samples are 'temporally synchronized,' but provides no reported synchronization method (e.g., hardware trigger details or NTP offsets), no measured time deltas, and no environmental metrics (turbidity readings, lux values) to confirm the claimed harsh conditions hold across the collection. All supported tasks (stereo matching, depth completion) depend on this fidelity.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript introduces ShotcreteDepth, a bi-modal dataset comprising 11,252 temporally synchronized stereo RGB images and LiDAR point clouds captured in shotcrete construction environments (including active shotcreting) under harsh conditions of high turbidity and poor illumination, along with 220 annotated samples and a lightweight LiDAR annotation tool. The dataset is positioned to support research in stereo matching, depth completion, and depth estimation for industrial robotic applications.","tokens_in":1812,"tokens_out":304,"duration_ms":13369,"significance":"Release of synchronized multi-modal sensor data from a real industrial construction domain, together with an annotation tool, would address a gap in publicly available datasets for perception under challenging conditions; if the synchronization and environmental fidelity claims are substantiated, the resource could enable targeted algorithm development and benchmarking for autonomous systems in similar settings.","major_comments":[{"comment":"Abstract: The central claims that the 11,252 samples are 'temporally synchronized' and were 'acquired under harsh real-world conditions, including high turbidity and poor illumination' are load-bearing for all stated use cases (stereo matching, depth completion, depth estimation), yet the manuscript provides no description of the synchronization hardware or protocol, no measured time-offset statistics, and no quantitative environmental metrics (turbidity, lux, or equivalent) to confirm the conditions across the collection.","section":"Abstract"}],"minor_comments":[],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive feedback on our manuscript. The recommendation for major revision is noted, and we address the concerns regarding substantiation of the synchronization and environmental condition claims below. We will revise the manuscript to strengthen these aspects where feasible.","responses":[{"response":"We agree that additional details are needed to support these claims. In the revised manuscript, we will expand the methods section to describe the synchronization hardware (including the specific trigger mechanism and cabling) and the protocol employed to achieve temporal alignment between the stereo RGB cameras and LiDAR sensor. We will also provide any available supporting information on the collection setup. However, time-offset statistics were not measured during acquisition, and quantitative environmental metrics such as turbidity or lux values were not recorded. We will explicitly note these limitations and enhance the qualitative description of the harsh conditions based on the operational context of active shotcreting.","revision_made":"partial","referee_comment":"[Abstract] Abstract: The central claims that the 11,252 samples are 'temporally synchronized' and were 'acquired under harsh real-world conditions, including high turbidity and poor illumination' are load-bearing for all stated use cases (stereo matching, depth completion, depth estimation), yet the manuscript provides no description of the synchronization hardware or protocol, no measured time-offset statistics, and no quantitative environmental metrics (turbidity, lux, or equivalent) to confirm the conditions across the collection."}],"tokens_in":1233,"tokens_out":366,"duration_ms":12034,"standing_objections":["Provision of measured time-offset statistics for synchronization, as these data were not collected during the original dataset acquisition.","Provision of quantitative environmental metrics (turbidity, lux, or equivalent), as these were not recorded during data collection."]},"desk_editor":{"model":"grok-4.3","letter":"The paper's main contribution is ShotcreteDepth, a collection of 11,252 stereo RGB and LiDAR pairs from active shotcrete sites and other construction areas. It also ships a simple annotation tool for the point clouds. That fills a narrow gap—most public depth datasets skip these dusty, poorly lit industrial settings—so the release itself is the new part.\n\nWhat works is the scale and the domain focus. Eleven thousand synchronized samples is enough to train or test stereo matching and depth completion models, and the 220 annotated frames give a starting point for evaluation. Releasing the tool lowers the barrier for others who want to add labels.\n\nThe soft spot is the missing verification. The abstract states the data were captured under high turbidity and poor illumination and that the pairs are temporally synchronized, yet it reports no lux readings, turbidity measurements, hardware trigger details, or measured time offsets. Without those, it is hard to know how faithfully the data match the claimed conditions. The small annotated set also means most users will still need to label a lot themselves.\n\nThis is for people already working on construction robotics or robust depth estimation who need domain-specific data. If the full paper or repo adds the collection logs and metrics, the dataset could support incremental experiments. I would send it to peer review because dataset papers in applied robotics often need referee input on documentation and utility even when the core claim is just the release.","headline":"This is a basic dataset release for depth perception in shotcrete and construction scenes, but the abstract gives no numbers or methods to back up the synchronization and harsh-condition claims.","tokens_in":2319,"tokens_out":365,"would_cite":false,"duration_ms":17353,"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":"ShotcreteDepth provides a bi-modal dataset of synchronized stereo RGB and LiDAR data from active shotcrete construction under high turbidity and poor illumination.","keywords":["ShotcreteDepth","dataset","depth estimation","stereo matching","LiDAR","construction robotics","depth completion","shotcrete"],"falsifier":"If depth estimation or completion models that succeed on the 220 annotated samples show no performance gain when deployed on independent recordings from the same shotcrete sites, the dataset's claimed representativeness would be refuted.","tokens_in":2632,"feed_emoji":"🏗","tokens_out":642,"duration_ms":24167,"temperature":0.7,"pith_summary":"The paper introduces ShotcreteDepth to supply realistic sensor data for testing robotic depth perception systems in construction settings. It collects 11,252 temporally synchronized stereo images and LiDAR point clouds from both active shotcreting processes and general sites, where turbidity and low light create noisy, incomplete measurements. The release includes a lightweight annotation tool for LiDAR and 220 annotated samples to support evaluation. This setup allows work on stereo matching, depth completion, and depth estimation that matches industrial operating conditions.","feed_headline":"Dataset supplies depth data from active shotcrete sites","feed_subtitle":"11,252 synchronized stereo and LiDAR samples target perception failures caused by turbidity and poor illumination in construction.","key_machinery":"The ShotcreteDepth bi-modal dataset of stereo RGB imagery paired with LiDAR point clouds collected under harsh construction conditions.","core_discovery":"The paper establishes the ShotcreteDepth dataset as a collection of 11,252 temporally synchronized stereo RGB images and LiDAR point clouds acquired in real-world shotcrete construction environments that feature high turbidity and poor illumination, accompanied by a lightweight annotation tool for LiDAR point clouds and 220 annotated samples for evaluation in stereo matching, depth completion, and depth estimation tasks.","pith_inferences":["The same data collection approach could be replicated for other dusty or low-visibility industrial tasks such as mining or tunneling.","Models pretrained on general outdoor datasets may require fine-tuning on this data to handle domain-specific noise patterns.","The dataset highlights sensor-fusion needs that could drive new hardware designs for construction robots.","Release of the annotation tool may accelerate labeling efforts in other point-cloud-heavy robotics domains."],"forward_implications":["Stereo matching algorithms can be evaluated on imagery degraded by construction turbidity.","Depth completion techniques gain a testbed for recovering structure from sparse, noisy LiDAR returns in low light.","Depth estimation research obtains examples that reflect the incomplete observations typical of industrial robotics.","Autonomous construction systems can be trained and validated against data that includes both active spraying and static site conditions.","The annotation tool enables rapid expansion of labeled point clouds for further experiments."],"fun_headline_variants":["Bi-modal ShotcreteDepth for robotic depth in construction","Stereo RGB and LiDAR data from turbid shotcrete sites","Dataset for depth estimation under poor illumination","11,252 synced samples with LiDAR annotation tool"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The collected stereo RGB imagery and LiDAR point clouds are temporally synchronized and accurately represent the high turbidity and poor illumination of active construction environments.","fun_headline_variants_meta":{"raw":{"variants":["Bi-modal ShotcreteDepth for robotic depth in construction","Stereo RGB and LiDAR data from turbid shotcrete sites","Dataset for depth estimation under poor illumination","11,252 synced samples with LiDAR annotation tool"]},"model":"grok-4.3","cost_usd":0.006382,"raw_usage":{"total_tokens":2873,"prompt_tokens":587,"num_sources_used":0,"completion_tokens":59,"cost_in_usd_ticks":63815500,"prompt_tokens_details":{"text_tokens":587,"audio_tokens":0,"image_tokens":0,"cached_tokens":64},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2227,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":587,"tokens_out":59,"duration_ms":14401,"temperature":1.0,"reasoning_tokens":2227,"cache_read_input_tokens":64,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-26T08:31:46.665269+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"If depth estimation or completion models that succeed on the 220 annotated samples show no performance gain when deployed on independent recordings from the same shotcrete sites, the dataset's claimed representativeness would be refuted.","supporting_citations":[],"review_version":1}