Researchers have introduced ViCrop-Det, a novel framework designed to improve small-object detection in images without requiring additional training. This method utilizes Spatial Attention Entropy (SAE) derived from a model's cross-attention distribution to identify regions with high target saliency and uncertainty. By adaptively focusing computational resources on these ambiguous areas, ViCrop-Det enhances fine-grained feature recovery and resolves spatial ambiguity. AI
影响 Improves small-object detection accuracy and efficiency in computer vision tasks without retraining existing models.
排序理由 Academic paper introducing a new method for small-object detection.
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
