Model Anomaly or Bbox Processing Error
giffmana · x · 2026-07-12
This reply analyzes potential causes for a multimodal model anomaly:
- If CoT (Chain of Thought) is used, the issue might be that cropped bbox coordinates were not synced properly.
- Another common scenario: random crop augmentation was applied during training, but the bboxes were not corrected accordingly.
- The responsibility likely lies in the preprocessing code or a specific slice of the dataset. The author hopes it's the former, as the latter would be much more troublesome.
More from Models
- Kimi K3 is praised for stronger English, frontend arena #1, and better handling of nuanced prompts — EXM7777 · 2026-07-22
- Gary Marcus says LLM math skills are like knowing only a car’s engine size — GaryMarcus · 2026-07-22
- OpenAI’s Codex + GPT-5.6 Sol hits 99% recall in Project APE verification tests — soumitrashukla9 · 2026-07-22
- OpenAI-linked paper says capability RL can make models more reward-seeking — MariusHobbhahn · 2026-07-22
- Macaron V1 adds LoRA RL on GLM 5.2 and claims SOTA benchmark gains — Xianbao_QIAN · 2026-07-22
- OpenAI rolls out voice in GPT-Live, but the UI obscures search and reasoning — Graham_dePenros · 2026-07-22