Tag: #히트맵

← All posts

• Human Pose 1: Finding Keypoints with Heatmaps — OpenPose's PAF, HRNet, and Top-Down vs. Bottom-Up

  We explain why pose estimation shifted from directly regressing coordinates (DeepPose) to drawing heatmaps for each keypoint, the two paths for handling multiple people (top-down which finds people first, and bottom-up w

  2026-10-17 · 16 min read #비전 모델#자세 추정#OpenPose#PAF#HRNet

• Loss Functions and Training Techniques: Cross-Entropy, Focal Loss, Heatmap Loss, Augmentation, Learning Rate Schedules, Mixed Precision

  We review loss functions that define what a model is considered to have done well, alongside techniques that make training fast and stable. We show with numbers how focal loss solves the problem of cross-entropy getting

  2026-10-17 · 14 min read #비전 모델#손실 함수#focal loss#데이터 증강#학습률

• U-Net and HRNet: Resolution-Preserving Architectures and Heatmap Output

  Classification models shrink images to figure out "what" is in them, but tasks like ball tracking, pose estimation, and segmentation require knowing "exactly where." We examine the design rationale of U-Net (2015), which

  2026-10-17 · 12 min read #비전 모델#U-Net#HRNet#히트맵#분할