summaryrefslogtreecommitdiff
path: root/training
AgeCommit message (Expand)Author
13 hoursfix: Compute gray from [0,1] RGB in CNN shader generatorskal
18 hoursfix: Complete auxiliary texture initialization fixskal
19 hoursadd --save-intermediates to train.py and cnn_testskal
20 hoursfix: Move sigmoid activation to call site in CNN layer shaderskal
20 hoursfix: Replace clamp with sigmoid in CNN final layerskal
20 hoursfeat: Add early stopping to CNN trainingskal
20 hoursfix: CNN training/inference to match WGSL sliding windowskal
21 hoursformat .wgsl layer code (cosmetics)skal
29 hoursfix: Use patch-based inference to match CNN training distributionskal
29 hoursopt: Move invariant in1 calculation outside CNN convolution loopsskal
31 hoursopt: Vec4-optimize CNN convolution shaders for SIMDskal
31 hourschore: Update CNN architecture to 3×3×3 with new trained weightsskal
31 hoursdocs: Update CNN training documentation with patch extractionskal
31 hoursfeat: Add salient-point patch extraction for CNN trainingskal
33 hoursfix: Correct UV coordinate computation to match PyTorch linspaceskal
33 hoursfix: Add clamp to CNN final layer to match PyTorch trainingskal
33 hoursrefactor: Optimize CNN grayscale computationskal
33 hoursupdate train_cnn.py and shaderskal
34 hoursfeat: Add inference mode to train_cnn.py for ground truth generationskal
34 hoursfix: CNN training normalization pipeline consistencyskal
35 hoursrefactor: Optimize CNN normalization to eliminate redundant conversionsskal
36 hoursfix: Support variable kernel sizes in CNN layer generationskal
37 hoursfeat: CNN RGBD→grayscale with 7-channel augmented inputskal
38 hoursudpateskal
41 hoursfeat: Add multi-layer CNN support with framebuffer capture and blend controlskal
43 hoursdocs: Update and streamline CNN training documentationskal
43 hoursfeat: Add checkpointing support to CNN training scriptskal
43 hoursfix: Auto-expand single kernel size to all layers in training scriptskal
43 hoursupdate target imagesskal
43 hoursfeat: Add coordinate-aware CNN layer 0 for position-dependent stylizationskal