Datasets:

Languages:
English
ArXiv:
License:
File size: 5,170 Bytes
234f610
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
"""Simple retrieval-style reference baseline for the image generation benchmark.



The baseline uses only the packaged metadata and retrieves the nearest training

example by RGB distance for every test sample.

"""

import argparse
import json
import math
from pathlib import Path


HF_ROOT = Path(__file__).resolve().parents[1]


def parse_args():
    parser = argparse.ArgumentParser(
        description="Run a nearest-RGB retrieval baseline on the image generation split."
    )
    parser.add_argument(
        "--hf-root",
        default=str(HF_ROOT),
        help="Path to the exported huggingface package root.",
    )
    parser.add_argument(
        "--output",
        help="Optional JSON output path for the retrieval summary.",
    )
    parser.add_argument(
        "--save-retrievals",
        help="Optional JSONL output path for per-sample retrieval results.",
    )
    parser.add_argument(
        "--compact",
        action="store_true",
        help="Write compact JSON instead of pretty JSON.",
    )
    return parser.parse_args()


def load_json(path: Path):
    with open(path, "r", encoding="utf-8") as handle:
        return json.load(handle)


def dump_json(value, compact: bool):
    if compact:
        return json.dumps(value, ensure_ascii=False, separators=(",", ":"))
    return json.dumps(value, ensure_ascii=False, indent=2)


def write_text(text: str, output_path: str | None):
    if output_path:
        path = Path(output_path)
        path.parent.mkdir(parents=True, exist_ok=True)
        with open(path, "w", encoding="utf-8") as handle:
            handle.write(text)
        return
    print(text)


def rgb_distance(left, right):
    return math.sqrt(
        (left[0] - right[0]) ** 2
        + (left[1] - right[1]) ** 2
        + (left[2] - right[2]) ** 2
    )


def nearest_neighbor(sample, train_samples):
    best_match = None
    best_distance = None
    for candidate in train_samples:
        distance = rgb_distance(sample["color_rgb"], candidate["color_rgb"])
        if best_distance is None or distance < best_distance:
            best_distance = distance
            best_match = candidate
    return best_match, best_distance


def safe_rate(correct: int, total: int):
    if total == 0:
        return None
    return correct / total


def main():
    args = parse_args()
    hf_root = Path(args.hf_root)

    train_metadata = load_json(hf_root / "image_generation" / "train" / "metadata.json")
    test_metadata = load_json(hf_root / "image_generation" / "test" / "metadata.json")

    train_samples = list(train_metadata.values())
    test_samples = list(test_metadata.values())

    retrievals = []
    total_distance = 0.0
    transparency_correct = 0
    transparency_total = 0
    surface_correct = 0
    surface_total = 0
    color_family_correct = 0
    color_family_total = 0

    for sample in test_samples:
        match, distance = nearest_neighbor(sample, train_samples)
        total_distance += distance
        retrieval = {
            "test_id": sample["id"],
            "retrieved_train_id": match["id"],
            "distance": distance,
            "test_image_path": sample["image_path"],
            "retrieved_image_path": match["image_path"],
        }
        retrievals.append(retrieval)

        if sample.get("transparency") is not None and match.get("transparency") is not None:
            transparency_total += 1
            transparency_correct += int(sample["transparency"] == match["transparency"])
        if sample.get("surface") is not None and match.get("surface") is not None:
            surface_total += 1
            surface_correct += int(sample["surface"] == match["surface"])
        if sample.get("color_family") is not None and match.get("color_family") is not None:
            color_family_total += 1
            color_family_correct += int(sample["color_family"] == match["color_family"])

    summary = {
        "benchmark": "image_generation",
        "baseline": "nearest_train_rgb_retrieval",
        "train_samples": len(train_samples),
        "test_samples": len(test_samples),
        "mean_retrieval_distance": total_distance / len(test_samples),
        "label_agreement": {
            "transparency": {
                "evaluated_samples": transparency_total,
                "accuracy": safe_rate(transparency_correct, transparency_total),
            },
            "surface": {
                "evaluated_samples": surface_total,
                "accuracy": safe_rate(surface_correct, surface_total),
            },
            "color_family": {
                "evaluated_samples": color_family_total,
                "accuracy": safe_rate(color_family_correct, color_family_total),
            },
        },
    }

    if args.save_retrievals:
        jsonl = "\n".join(json.dumps(item, ensure_ascii=False) for item in retrievals)
        write_text(jsonl, args.save_retrievals)

    write_text(dump_json(summary, args.compact), args.output)


if __name__ == "__main__":
    main()