Unverified50% confidenceSolutionExact time
For a small dataset of 500 images, the recommendation is to first manually annotate 80-100 images precisely as a seed set, fine-tune a lightweight detector with the seed set, then pre-annotate the remaining 400 images, take the union of the two rounds of pre-annotation results, and have humans confirm and correct them.
1
Sources
50%
Confidence
Long-term
Relevance
10/20/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe paper constructs a diagnostic experiment using one-step pushing, validated with 32 random seeds.68% similarUnverified作者建议的组合策略是用至少5个随机种子分别训练基线和提出方法，报告准确率均值和标准差配合配对检验，并对代表性种子额外运行 McNemar 检验展示样本级预测分歧67% similarUnverifiedThe researchers conducted a full end-to-end test on a set of 500 images, with bounding boxes automatically generated by the detector rather than using manually annotated boxes.66% similarUnverified通过固定随机种子（Seed）可以提升测试的确定性66% similarUnverifiedThe researchers calibrated the Agentic Judge using 100 trajectories that were manually annotated call by call.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1075021API
curl https://kongchang.com/api/v1/knowledge/claims/1075021MCP
get_claim(id=1075021)