flashvtg-experiment-backup / FlashVTG /scripts /find_similar_pair_interference_examples.py
zhaoshiwen's picture
Add files using upload-large-folder tool
57259ac verified
Raw
History Blame Contribute Delete
6.5 kB
#!/usr/bin/env python
"""
从验证集中找出与 qid 158 同类型的 10 个例子:
- 有多组 (noun, verb) pair 形成干扰(query 描述多个可能时刻)
- 只有一组是 GT
- 模型 top-1 预测正确(IoU >= 0.5)
"""
import json
import argparse
import re
from pathlib import Path
def load_jsonl(path):
out = []
with open(path, "r", encoding="utf-8") as f:
for line in f:
line = line.strip()
if not line:
continue
out.append(json.loads(line))
return out
def iou_segment(pred, gt):
"""pred/gt: [start, end] in seconds. Returns IoU."""
p_s, p_e = float(pred[0]), float(pred[1])
g_s, g_e = float(gt[0]), float(gt[1])
inter_s = max(p_s, g_s)
inter_e = min(p_e, g_e)
inter = max(0, inter_e - inter_s)
union = (p_e - p_s) + (g_e - g_s) - inter
return inter / union if union > 0 else 0.0
def has_multiple_pairs_interference(query):
"""
启发式:query 是否描述多组事件/多组 pair,形成干扰。
- 含 before / after / and / then / while / or 等多事件连接
- 逗号分隔多动作、多名词多动词结构
"""
q = (query or "").lower()
# 多事件连接词(与 qid158 "before" 同型)
if re.search(r"\b(before|after|and then|then\b|while\b|whilst| or )\b", q):
return True
# "X and Y" 结构(多主体/多动作)
if re.search(r"\band\b", q) and len(q.split()) >= 6:
return True
# 逗号分隔多动作
if q.count(",") >= 1 and (q.count("ing ") >= 2 or " and " in q):
return True
# 较长句且含多个动词/名词线索
words = q.split()
if len(words) >= 10 and (" is " in q or " are " in q or "ing " in q):
return True
return False
def main():
parser = argparse.ArgumentParser(description="Find 10 examples similar to qid 158 (multi-pair, model correct)")
parser.add_argument("--val_jsonl", type=str, default="data/highlight_val_release.jsonl", help="Val GT jsonl")
parser.add_argument("--pred_jsonl", type=str, default=None, help="Pred jsonl (nms). Default: results/.../best_*_nms_thd_0.7.jsonl")
parser.add_argument("--results_dir", type=str, default="results/qv_internvideo2-video_tef-baseline_strict-2026-02-21-18-59-41", help="Results dir to find pred file")
parser.add_argument("--ref_qid", type=int, default=158, help="Reference qid (same type)")
parser.add_argument("--topk", type=int, default=10, help="Number of examples to output")
parser.add_argument("--iou_threshold", type=float, default=0.5, help="Min IoU for model correct")
args = parser.parse_args()
root = Path(__file__).resolve().parent.parent
val_path = root / args.val_jsonl
if not val_path.exists():
val_path = Path(args.val_jsonl)
if not val_path.exists():
raise FileNotFoundError(f"Val jsonl not found: {val_path}")
pred_path = None
if args.pred_jsonl:
pred_path = Path(args.pred_jsonl)
else:
res_dir = root / args.results_dir
cand = list(res_dir.glob("best_*_val_preds_nms_thd_0.7.jsonl"))
if cand:
pred_path = cand[0]
if not pred_path or not pred_path.exists():
raise FileNotFoundError(f"Pred jsonl not found: {pred_path}")
val_list = load_jsonl(val_path)
pred_list = load_jsonl(pred_path)
val_by_qid = {item["qid"]: item for item in val_list}
pred_by_qid = {item["qid"]: item for item in pred_list}
# 收集:多 pair 干扰 + 模型 top-1 正确
candidates = []
for qid, gt_item in val_by_qid.items():
if qid not in pred_by_qid:
continue
pred_item = pred_by_qid[qid]
query = gt_item.get("query", "")
gt_windows = gt_item.get("relevant_windows", [])
if not gt_windows:
continue
pred_windows = pred_item.get("pred_relevant_windows", [])
if not pred_windows:
continue
top1 = pred_windows[0]
pred_seg = [top1[0], top1[1]]
max_iou = max(iou_segment(pred_seg, gw) for gw in gt_windows)
if max_iou < args.iou_threshold:
continue
if not has_multiple_pairs_interference(query):
continue
candidates.append({
"qid": qid,
"query": query,
"vid": gt_item.get("vid", ""),
"relevant_windows": gt_windows,
"pred_top1": pred_seg,
"pred_score": top1[2] if len(top1) >= 3 else None,
"max_iou": max_iou,
})
# 优先保留 ref_qid,再按与 ref 的相似度(同有 before / and 等)排序,再按 iou 降序
ref = next((c for c in candidates if c["qid"] == args.ref_qid), None)
ref_query = (ref or {}).get("query", "").lower()
def score_similarity(c):
q = c["query"].lower()
s = 0
if "before" in ref_query and "before" in q:
s += 2
if " and " in ref_query and " and " in q:
s += 1
if " then " in ref_query or " then " in q:
s += 0.5
return (s, c["max_iou"])
candidates.sort(key=lambda c: (c["qid"] != args.ref_qid, -score_similarity(c)[0], -score_similarity(c)[1]))
# 确保 ref_qid 在列首(若在候选里)
out = []
for c in candidates:
if c["qid"] == args.ref_qid:
out.insert(0, c)
else:
out.append(c)
# 去重并保持顺序
seen = set()
unique = []
for c in out:
if c["qid"] in seen:
continue
seen.add(c["qid"])
unique.append(c)
selected = unique[: args.topk]
print(f"# Found {len(candidates)} candidates (multi-pair + model R1 correct @ IoU>={args.iou_threshold}). Selected {len(selected)} (ref_qid={args.ref_qid}):\n")
for i, c in enumerate(selected, 1):
mark = " <-- ref" if c["qid"] == args.ref_qid else ""
print(f"{i}. qid={c['qid']}{mark}")
print(f" query: {c['query']}")
print(f" vid: {c['vid']}")
print(f" GT: {c['relevant_windows']} | pred_top1: {c['pred_top1']} (score={c['pred_score']:.4f}) IoU={c['max_iou']:.3f}")
print()
out_json = root / "results" / "similar_to_qid158_examples.json"
out_json.parent.mkdir(parents=True, exist_ok=True)
with open(out_json, "w", encoding="utf-8") as f:
json.dump(selected, f, indent=2, ensure_ascii=False)
print(f"Saved {len(selected)} examples to {out_json}")
if __name__ == "__main__":
main()