=========================================================================== TrackChain — Enhanced PatchCore Visual Anomaly Detector (tc.v1 SOTA) =========================================================================== Backbone: wide_resnet50_2 Normal Dataset: /home/iedc_ai_dgx1/TrackChain/data/external/rail_normal_expanded Patch Size: 3x3 (Multi-scale Layer2 + Layer3) Coreset Ratio: 10.0% (Max Coreset: 3000) Batch Size: 32 Calib P99 Target:99.0% Compute Device: cuda [1/5] Loaded 779 normal training images, 206 normal validation images. [2/5] Extracting multi-scale patch embeddings (Layer2 + Layer3) from normal track... Extracting normal patch embeddings: 0%| | 0/25 [00:00 3000 representative patches)... Minimax Coreset Selection: 0%| | 0/3000 [00:00