{ "benchmarks": { "semantic/cityscapes": { "capability": "2d", "dataset_key": "semantic", "root": "data/raw/cityscapes", "split": "val", "native_resolution": true, "dataset_kwargs": {}, "samples": 500, "source_samples": 500, "dataset_fingerprint": "4800afab564d03887787292933196637cb183d48e4460eae3e71e2d3056649e3", "paper_reference": { "benchmark": "semantic", "dataset": "cityscapes", "split": "val", "metric": "mIoU", "paper": 0.699, "note": "Cityscapes val, 19-class legend including absent classes" }, "decoded_metrics": { "mIoU": 0.21488772332668304, "pixel_accuracy": 0.6165922528089834, "per_class_iou": [ 0.9111851453781128, 0.5919085144996643, 0.0002911773626692593, 0.0245357658714056, 0.14803653955459595, 0.030759338289499283, 0.0013343992177397013, 0.32152560353279114, 0.8069646954536438, 0.18592989444732666, 0.8710127472877502, 0.03289300203323364, 0.15516267716884613, 2.678571945580188e-06, 0.0, 0.0, 9.742333531903569e-06, 0.0, 0.0013146874262019992 ], "valid_pixels": 917018489 }, "merged_shards": { "count": 2, "blocks": [ [ 0, 250 ], [ 250, 500 ] ] } }, "depth/diode": { "capability": "3d", "dataset_key": "diode", "root": "data/raw/diode", "split": "val", "native_resolution": true, "dataset_kwargs": { "scene_type": "indoors" }, "samples": 325, "source_samples": 325, "dataset_fingerprint": "232d580aad6e70e832582574952aae2881d5446df76700a65263699e67b4f7b1", "paper_reference": { "benchmark": "depth", "dataset": "diode", "metric": "delta_1.25", "paper": 0.917, "paper_absrel": 0.108, "note": "DIODE-Indoor" }, "decoded_metrics": { "AbsRel": 2369.7988923507623, "SqRel": 2298440481.7564754, "RMSE": 86640.62652330476, "RMSE_log": 2.0387466503864773, "delta_1.25": 0.07701537999848698, "delta_1.25^2": 0.1616714156651281, "delta_1.25^3": 0.26458824917186297, "valid_pixels": 251430909, "AbsRel_per_image": 2510.145138869721, "RMSE_per_image": 24996.572537966968, "delta_1.25_per_image": 0.07618638903987175, "images": 325, "AbsRel_aligned": 884.0034807925355, "delta_1.25_aligned": 0.5702212775305245, "RMSE_aligned": 121488.23251882072, "aligned_valid_pixels": 250644970, "skipped_samples": 0, "depth_curve": "metric_power", "depth_units": "metres" } }, "depth/eth3d": { "capability": "3d", "dataset_key": "eth3d", "root": "data/raw/eth3d", "split": "val", "native_resolution": true, "dataset_kwargs": {}, "samples": 454, "source_samples": 454, "dataset_fingerprint": "fe5358029536a5dcf622d0263432d53bb5ec737d4805b5078c06760748a840e2", "paper_reference": { "benchmark": "depth", "dataset": "eth3d", "metric": "delta_1.25", "paper": 0.935, "paper_absrel": 0.103 }, "decoded_metrics": { "AbsRel": 2.5183310746559884, "SqRel": 79.38748302880606, "RMSE": 17.986439332820574, "RMSE_log": 1.9709849963989834, "delta_1.25": 0.11925263062829346, "delta_1.25^2": 0.23443484210639565, "delta_1.25^3": 0.33277493130946484, "valid_pixels": 127712646, "AbsRel_per_image": 2.5737888975651515, "RMSE_per_image": 12.462342106002806, "delta_1.25_per_image": 0.1285937691920154, "images": 454, "AbsRel_aligned": 1.9480595149436388, "delta_1.25_aligned": 0.7040436249285035, "RMSE_aligned": 134.9028541449569, "aligned_valid_pixels": 124336433, "skipped_samples": 0, "depth_curve": "metric_power", "depth_units": "metres" }, "merged_shards": { "count": 2, "blocks": [ [ 0, 227 ], [ 227, 454 ] ] } }, "depth/nyu": { "capability": "3d", "dataset_key": "nyu_depth", "root": "data/raw/nyu_depth_v2", "split": "val", "native_resolution": true, "dataset_kwargs": {}, "samples": 654, "source_samples": 654, "dataset_fingerprint": "7b97e1125ad9f4322e0bec7a2c6a354dc809ab2f3ed2451b2c68d9c8f7ecb953", "paper_reference": { "benchmark": "depth", "dataset": "nyu", "metric": "delta_1.25", "paper": 0.948, "paper_absrel": 0.081 }, "decoded_metrics": { "AbsRel": 1.722833826843339, "SqRel": 9.855502423959873, "RMSE": 4.574520489150392, "RMSE_log": 1.6363381378304396, "delta_1.25": 0.0723108977514424, "delta_1.25^2": 0.16879996851650167, "delta_1.25^3": 0.30323655105967096, "valid_pixels": 156018240, "AbsRel_per_image": 1.7228337658928075, "RMSE_per_image": 4.180940210636967, "delta_1.25_per_image": 0.07231089552970879, "images": 654, "AbsRel_aligned": 2.283583821095263, "delta_1.25_aligned": 0.6566354722876154, "RMSE_aligned": 24.507983961218898, "aligned_valid_pixels": 154348320, "skipped_samples": 0, "depth_curve": "metric_power", "depth_units": "metres" }, "merged_shards": { "count": 2, "blocks": [ [ 0, 327 ], [ 327, 654 ] ] } }, "referring/reasonseg": { "capability": "2d", "dataset_key": "reasonseg", "root": "data/raw/reasonseg", "split": "val", "native_resolution": true, "dataset_kwargs": {}, "samples": 200, "source_samples": 200, "dataset_fingerprint": "0506037a629f6ca966100ce63f2aa1df9d56e192228e9741cbb0c9d5402fac84", "paper_reference": { "benchmark": "referring", "dataset": "reasonseg", "split": "val", "metric": "gIoU", "paper": 0.793, "note": "ReasonSeg val; paper rewrites queries with Gemini 2.5 Pro" }, "decoded_metrics": { "IoU": 0.2543628777739729, "cIoU": 0.1731907588495319, "gIoU": 0.2543628777739729, "Precision@0.5": 0.18, "Precision": 0.17935487757824095, "Recall": 0.8344168456960918, "F-score": 0.29524739696955726, "samples": 200 } }, "referring/refcocog": { "capability": "2d", "dataset_key": "referring", "root": "data/raw/refer/refcocog", "split": "val", "native_resolution": true, "dataset_kwargs": { "protocol": "umd", "strict": true }, "samples": 4896, "source_samples": 4896, "dataset_fingerprint": "e570d18c8f6517ec0a39616a19a91c568ca7f8b655055180a462c6e28fe5b298", "paper_reference": { "benchmark": "referring", "dataset": "refcocog", "split": "val", "metric": "cIoU", "paper": 0.738, "note": "RefCOCOg UMD val" }, "decoded_metrics": { "IoU": 0.5253263977825223, "cIoU": 0.37619617476788464, "gIoU": 0.5253263977825223, "Precision@0.5": 0.5526960784313726, "Precision": 0.39575273335307415, "Recall": 0.8838940809294463, "F-score": 0.5467188205654409, "samples": 4896 }, "merged_shards": { "count": 10, "blocks": [ [ 0, 490 ], [ 490, 980 ], [ 980, 1470 ], [ 1470, 1960 ], [ 1960, 2450 ], [ 2450, 2940 ], [ 2940, 3430 ], [ 3430, 3920 ], [ 3920, 4410 ], [ 4410, 4896 ] ] } } } }