File size: 1,860 Bytes
4bbb0db b78323b 4bbb0db b78323b 4bbb0db | 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 | {
"app": "neat-change-studio",
"version": 1,
"models": {
"bf16": {
"title": "CYWS bf16",
"about": "The most faithful build: matches the FP32 model's boxes one for one. The default.",
"path": "bf16",
"precision": "BF16",
"latency_ms": 35.2,
"accuracy": {
"changes_found": "33 / 47",
"pixel_f1": 0.587,
"false_alarm_pairs": "2 / 15",
"box_agreement_vs_fp32": 1.0,
"heat_cosine_min": 0.994
},
"card": {
"model": "The Change You Want to See (Sachdeva & Zisserman, WACV 2023)",
"architecture": "Siamese ResNet-50 U-Net, co-attention, CenterNet head; 49.5M parameters",
"input": "before + after, 256x256 RGB each",
"upstream": "ragavsachdeva/The-Change-You-Want-to-See",
"license": "Apache-2.0"
},
"files": [
{
"name": "cyws_bf16_mpk.tar.gz",
"bytes": 119178396,
"sha256": "ba5a5a12a9858dd31f265d338e55c1cdd7df5dd80a1b96a64d9763d28ed8c444"
}
]
},
"w8a16": {
"title": "CYWS w8a16",
"about": "Int8 weights with bf16 activations: 40% smaller and ~23% faster per window, slightly less exact.",
"path": "w8a16",
"precision": "A_BF16_W_INT8",
"latency_ms": 27.2,
"accuracy": {
"changes_found": "31 / 47",
"pixel_f1": 0.581,
"false_alarm_pairs": "3 / 15",
"box_agreement_vs_fp32": 0.966,
"heat_cosine_min": 0.826
},
"card": {
"model": "The Change You Want to See (Sachdeva & Zisserman, WACV 2023)",
"architecture": "Siamese ResNet-50 U-Net, co-attention, CenterNet head; 49.5M parameters",
"input": "before + after, 256x256 RGB each",
"upstream": "ragavsachdeva/The-Change-You-Want-to-See",
"license": "Apache-2.0"
},
"files": [
{
"name": "cyws_w8a16_mpk.tar.gz",
"bytes": 71785644,
"sha256": "5d33702c13731558eb4fd8f71ca3c18d2b6fa0216d66931e3a31b743d43f0c78"
}
]
}
}
} |