{"claim":{"id":"54138bf7-016e-4268-82b7-7ba90f44b8fe","submission_id":"b0822c53-ac9d-4f29-95c5-cff5f4a5fe28","submission_revision_id":"bf4f55a7-1fc3-4a51-a53e-130fb14d5fac","builder":"builder-01","status":"active","claimed_at":"2026-10-15T14:40:48.264780Z","delivered_at":null,"ended_at":null,"delivery_note":null,"end_reason":null},"task":{"submission_id":"b0822c53-ac9d-4f29-95c5-cff5f4a5fe28","submission_revision_id":"bf4f55a7-1fc3-4a51-a53e-130fb14d5fac","revision_version":1,"submission_status":"under_review","submitted_at":"2026-07-09T22:58:19.810980Z","submitter_name":"Nikola Zubic and Davide Scaramuzza","submitter_email":"zubic@ifi.uzh.ch","task":{"status":"submitted","version":1,"extra_fields":{},"submitted_at":"2026-07-09T22:58:19.810980+00:00","industry_domain":"Robotics & Autonomous Systems — Event-Based Motion Estimation","task_short_name":"event_vision_dense_flow_instance_1","operating_system":"Linux","task_description":"Objective: Estimate dense forward optical flow from an asynchronous event-camera recording containing camera-induced background motion, three independently moving foreground objects, occlusions, background activity, timestamp jitter, and persistent hot pixels.\n\nStarting State: The input directory contains:\n* events.h5: a one-second, 240×180 event-camera stream with /events/x (uint16), /events/y (uint16), /events/t_us (int64), and /events/p (int8, exactly -1 or +1).\n* query_timestamps.csv: 45 strictly increasing timestamps under the column timestamp_us.\n* calibration.yaml: image dimensions, pixel coordinates, polarity convention, support interval, flow direction, channel order, output units, and output schema.\n* README_INPUT.md: complete instructions and file conventions.\nNo intensity/RGB/APS frames, flow fields, visibility masks, object trajectories, layer masks, or motion parameters are supplied.\n\nInstructions and Rules:\nFor each query timestamp t, estimate forward flow over the half-open interval [t-10000,t+10000) microseconds. Channel order is [dx,dy]. Positive dx is rightward and positive dy is downward. Units are pixels per 20 ms interval, not pixels per second. Produce event_flow.h5 containing exactly:\n* /flow: little-endian float32, shape (45,180,240,2)\n* /timestamps_us: little-endian int64, shape (45,)\n/timestamps_us must exactly match query_timestamps.csv. Supply a finite flow vector at every pixel. Every absolute flow component must be at most 100 pixels. Store both datasets directly in the submitted HDF5 file. Virtual datasets, soft or external links, and external raw storage are prohibited. The complete event stream may be used, but no data outside the input package may be used. Do not use intensity/RGB/APS frames, pretrained models, downloaded data, network access, reference outputs, hidden masks, or hard-coded flow. Do not modify the input files.\nThe solution must be deterministic and runnable CPU-only with:\nbash run.sh <input_directory> <output_directory>\n\nRequired Deliverables:\n* event_flow.h5\n* run.sh\n* All source code required by run.sh\n* A pinned dependency manifest","software_licensing":"Free/Open Source","evaluation_criteria":null,"verification_method":"Run the submission twice in separate output directories and evaluate both runs using the supplied offline grader. A hidden reference validity mask is used, with an additional four-pixel image border excluded.\n\nFor candidate flow `u=(dx,dy)` and reference flow `u*`, endpoint error is:\nEPE = sqrt((dx-dx*)² + (dy-dy*)²)\n\nAll of the following gates are mandatory:\n1. The output filename, HDF5 datasets, dtypes, shapes, and timestamps must exactly match the specification.\n2. Virtual datasets, links, and external raw storage are rejected.\n3. All flow values must be finite and every absolute component must be at most `100 px`.\n4. Global mean EPE must be no greater than `0.80 px`.\n5. Global median EPE must be no greater than `0.45 px`.\n6. Global 95th-percentile EPE must be no greater than `2.50 px`.\n7. The fraction of evaluated pixels with `EPE > 3.0 px` must be no greater than `5.0%`.\n8. Mean augmented angular error must be no greater than `6.0°`, using:\nacos((u·u*+1)/(sqrt(||u||²+1)×sqrt(||u*||²+1)))\n9. Mean EPE at every individual timestamp must be no greater than `1.50 px`.\n10. On evaluated pixels where reference-flow magnitude is at least `1.50 px`, mean EPE must be no greater than `1.25 px`.\n 11. Two independently generated outputs must differ by no more than `1×10^-6` at any flow element, and timestamps must be exactly equal. The same physical file cannot be submitted as both executions.\n\nThreshold comparisons are inclusive. Metrics are calculated in `float64` without rescaling, interpolation, temporal alignment, manual cropping, or subjective review.","software_and_version":"Python 3.11.9; NumPy 2.0.1; SciPy 1.14.1; OpenCV-Python-Headless 4.10.0.84; h5py 3.11.0; PyYAML 6.0.2","specific_requirements":null,"expected_completion_time":null},"files":[{"file_id":"14cd37f3-a1cc-462a-a448-29746f015ac7","filename":"event_flow_input_v1.zip","category":"input_materials","content_type":"application/zip","size_bytes":464741},{"file_id":"650bf3a0-ebfc-4eb9-97f9-60061a496769","filename":"event_flow_reference_v1.zip","category":"reference_output","content_type":"application/zip","size_bytes":22003156}],"deep_review":{"overall_verdict":"Accept","review_text":"**Overall Verdict:** Accept — The submission is complete, coherent, objectively gradable, and operationally reachable, with only a reviewer-environment limitation preventing independent decoding of numerical HDF5 payloads.\n\n**Completeness Check:**\n- Instruction: Complete — The objective, event and flow conventions, temporal support, restrictions, deterministic runtime interface, deliverables, and exact output schema are consistently specified.\n- Inputs: Complete — The input package contains the event stream, 45-row timestamp CSV, calibration, schema, documentation, pinned dependencies, license, and a schema validator; provenance hashes for the core inputs agree across packages.\n- Toolchain: Complete — The CPU-only Python toolchain and solver dependencies are pinned, legally obtainable, and appropriate for the modest data scale; evaluator dependencies are separately pinned.\n- Reference output: Complete — The protected package contains reference flow and validity data, a gold-format candidate, provenance, diagnostics, and an offline evaluator. Its recorded gold self-test passes all nine grouped mandatory gates.\n- Rubric: Complete — The evaluator implements the documented schema, storage, range, EPE, angular, per-timestamp, high-motion, border-exclusion, and two-run determinism requirements. The zero-flow diagnostic fails four accuracy gates, demonstrating meaningful discrimination.\n\n**File and Task Reachability:**\nThe attachment set is `partially_reachable` solely because the review environment lacked an HDF5 numerical decoder and video decoder. Both `event_flow_input_v1.zip` and `event_flow_reference_v1.zip` were inspected at their manifest byte counts; they are valid, unencrypted ZIP archives with safe paths, no duplicate entries or special filesystem objects, modest expansion, and successful internal integrity checks. Text, CSV, JSON, YAML, licenses, evaluator source/tests, generator provenance, archive inventories, HDF5 signatures and embedded dataset names, and the MP4 container structure were inspected. Numerical HDF5 arrays and diagnostic video frames were not independently decoded; consistent generator source, cross-package hashes, evaluator diagnostics, and embedded schema names mitigate this reviewer-tooling uncertainty.\n\nThe task is `reachable`: a solver can load the supplied asynchronous events, generate two deterministic `event_flow.h5` outputs using the declared CPU-only toolchain, and be evaluated offline against the separately packaged hidden reference and validity mask. Storage and memory requirements are modest, reference material is separated from solver-visible inputs, and no external service or hardware dependency is present. License keys, accounts, tokens, and other credentials are not required from the contributor. Confidential, proprietary, or internal material may be evaluated as submitted evidence without requiring additional disclosure.\n\n**Special Screening:**\nNone — provenance affirmatively describes deterministic procedural generation without pretrained models or external assets. No evidence supports AI-generated core evidence, a public benchmark, or a synthetic defect fixture; public identifiability was not independently searched under the offline evidence boundary.\n\n**Recommended Changes:**\nThe submission provides unusually strong documentation, provenance, integrity manifests, negative and positive diagnostics, and evaluator tests. No contributor changes are required. An admin-side execution of the supplied evaluator tests and diagnostics in the declared Python/HDF5 environment would close the remaining reviewer-tooling uncertainty.","reviewed_at":"2026-08-21T16:18:59.571126Z"},"claim":{"id":"54138bf7-016e-4268-82b7-7ba90f44b8fe","submission_id":"b0822c53-ac9d-4f29-95c5-cff5f4a5fe28","submission_revision_id":"bf4f55a7-1fc3-4a51-a53e-130fb14d5fac","builder":"builder-01","status":"active","claimed_at":"2026-10-15T14:40:48.264780Z","delivered_at":null,"ended_at":null,"delivery_note":null,"end_reason":null}},"revision_is_current":true}