index
:
demo.git
main
Vide-coded 64k demo system
summary
refs
log
tree
commit
diff
log msg
author
committer
range
path:
root
/
cnn_v3
/
training
/
train_cnn_v3.py
Age
Commit message (
Expand
)
Author
15 hours
fix(cnn_v3): L1 loss + depth-grad tanh normalization to reduce flat convergence
HEAD
main
skal
15 hours
fix(cnn_v3): remove dec0 ReLU, load FiLM MLP at runtime
skal
40 hours
fix(cnn_v3/training): fix defaults and help strings across py tools
skal
40 hours
feat(cnn_v3): upgrade architecture to enc_channels=[8,16]
skal
2 days
fix(cnn_v3/training): rebuild optimizer before loading state on resume past F...
skal
3 days
feat(cnn_v3): 3×3 dilated bottleneck + Sobel loss + FiLM warmup + architectu...
skal
3 days
feat(cnn_v3/training): add --single-sample option + doc fixes
skal
5 days
feat(cnn_v3): patch alignment search, resume, Ctrl-C save
skal
5 days
docs(cnn_v3): add uv inline deps to train_cnn_v3.py + HOW_TO_CNN note
skal
6 days
fix(cnn_v3): correct weight budget in docstring (3.9→5.4 KB f16)
skal
7 days
feat(cnn_v3): Phase 6 — training script (train_cnn_v3.py + cnn_v3_utils.py)
skal
