summaryrefslogtreecommitdiff
path: root/training
AgeCommit message (Expand)Author
23 hoursfix: Complete auxiliary texture initialization fixskal
24 hoursadd --save-intermediates to train.py and cnn_testskal
25 hoursfix: Move sigmoid activation to call site in CNN layer shaderskal
25 hoursfix: Replace clamp with sigmoid in CNN final layerskal
25 hoursfeat: Add early stopping to CNN trainingskal
26 hoursfix: CNN training/inference to match WGSL sliding windowskal
26 hoursformat .wgsl layer code (cosmetics)skal
35 hoursfix: Use patch-based inference to match CNN training distributionskal
35 hoursopt: Move invariant in1 calculation outside CNN convolution loopsskal
36 hoursopt: Vec4-optimize CNN convolution shaders for SIMDskal
36 hourschore: Update CNN architecture to 3×3×3 with new trained weightsskal
36 hoursdocs: Update CNN training documentation with patch extractionskal
36 hoursfeat: Add salient-point patch extraction for CNN trainingskal
38 hoursfix: Correct UV coordinate computation to match PyTorch linspaceskal
38 hoursfix: Add clamp to CNN final layer to match PyTorch trainingskal
38 hoursrefactor: Optimize CNN grayscale computationskal
38 hoursupdate train_cnn.py and shaderskal
39 hoursfeat: Add inference mode to train_cnn.py for ground truth generationskal
39 hoursfix: CNN training normalization pipeline consistencyskal
40 hoursrefactor: Optimize CNN normalization to eliminate redundant conversionsskal
41 hoursfix: Support variable kernel sizes in CNN layer generationskal
42 hoursfeat: CNN RGBD→grayscale with 7-channel augmented inputskal
43 hoursudpateskal
46 hoursfeat: Add multi-layer CNN support with framebuffer capture and blend controlskal
2 daysdocs: Update and streamline CNN training documentationskal
2 daysfeat: Add checkpointing support to CNN training scriptskal
2 daysfix: Auto-expand single kernel size to all layers in training scriptskal
2 daysupdate target imagesskal
2 daysfeat: Add coordinate-aware CNN layer 0 for position-dependent stylizationskal