File size: 2,974 Bytes
d11c1ec | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 | {
"examples": 896,
"source_video_count": 461,
"qa_video_count": 461,
"source_hashes": {
"/home/jovyan/data-1/jusang/superdyn/motion_audit/outputs/temporal_geometry_fullscope/qa/final/qa.json": "4ed3eb2356fb055c6a5fdad1ba89ab233af2941dd343c4cbffe9f7bd9bb503ce"
},
"presentation": "All nine measurement RGB frames as ordered images, with frame labels and requested target contours only.",
"targets": "Original QA A/B mapping retained; model receives only visual marks and original question/descriptions.",
"excluded_from_model_input": [
"reference answer",
"answer index",
"depth",
"camera pose",
"geometry evidence",
"threshold-derived metrics"
],
"generation": {
"do_sample": false,
"num_beams": 1,
"max_new_tokens": 64
},
"group4_mapping": {
"LLL": "LLL",
"LLH": "LLH",
"HLL": "HLL+LHL+HHL",
"LHL": "HLL+LHL+HHL",
"HHL": "HLL+LHL+HHL",
"HLH": "HLH+LHH+HHH",
"LHH": "HLH+LHH+HHH",
"HHH": "HLH+LHH+HHH"
},
"group4_counts": {
"HLH+LHH+HHH": 896
},
"scoring": "Deterministic MCQ matching; invalid/error predictions count as incorrect; denominator 896 per model.",
"reference_status": "Provisional model-estimated geometry labels; no human review.",
"model_input_policy": "Same marked images, question and option order for all models; native per-model visual preprocessing.",
"scope": "Initial evaluation at nine frames; not the papers original benchmark protocol or a 16/32-frame setting.",
"jobs": [
{
"model": "qwen25vl_3b",
"gpu": 0,
"shard": 0,
"shards": 1
},
{
"model": "qwen25vl_7b",
"gpu": 1,
"shard": 0,
"shards": 1
},
{
"model": "vgllm_qa_4b",
"gpu": 2,
"shard": 0,
"shards": 1
},
{
"model": "vgllm_qa_8b",
"gpu": 3,
"shard": 0,
"shards": 2
},
{
"model": "vgllm_qa_8b",
"gpu": 4,
"shard": 1,
"shards": 2
},
{
"model": "mllm4d_sft_8b",
"gpu": 5,
"shard": 0,
"shards": 1
},
{
"model": "4drgpt_8b",
"gpu": 6,
"shard": 0,
"shards": 1
},
{
"model": "dsr_suite_7b",
"gpu": 7,
"shard": 0,
"shards": 1
}
],
"created_at": "2026-10-21T09:51:25.961939+00:00",
"native_input_packaging": {
"default": "nine ordered marked RGB images",
"dsr_suite_7b": "Same nine marked RGB frames packaged as a video; native final-frame temporal padding and 32 GSM query tokens. fps=2 is an index encoding convention, not measured seconds."
},
"excluded_sources": [
"dynamic_replica",
"dynpose-100k",
"point_odyssey",
"spring"
],
"additional_model_source": "https://github.com/TencentARC/DSR_Suite",
"selection_rule": "QA audit decision=keep AND group in HLH/LHH/HHH (N=L)",
"reused_evaluation": "/home/jovyan/data-1/jusang/superdyn/motion_audit/outputs/temporal_vlm_eval_fullscope"
} |