Yuhe Zhang

Is there a bug in dump_tensorflow_weights.py for the first box predictor layer ?

2

 Only the first box's scale is changed from 0.1 to 0.2 when switching from tf to caffe. But the script change the weights of all 3 boxes. Is this...

ppmatting模型可以convert成功，但无法用paddle.js load，error：unregisteredop equal, fill_constant

2

你好，我用paddlejsconverter转换官方的ppmatting model (hrnet_w18) 可以成功，但在paddle.js里面load会报错： Error: [unregistered op] fill_constant Error: [unregistered op] equal 请问能否支持这两个op？或者能否指导下怎么用registerOp?

Can you release model with smaller input size?

2

The input size 1024x1024 uses too much memory for my case, would you consider release models with smaller input size, such as 512x512? Or directly convert your pretrained model (pytorch?)...

L_con tends to make Z_i and Z_j very small, as the loss will decrease this way. But in your paper, the visualized Z_i and Z_j histogram is ranged well within...

feat: callback support #805

1

Implements PyTorch Lightning-style callback system for Automodel recipes to enable custom integrations (e.g., Customizer metrics reporting, custom monitoring). ### Key Features - **6 lifecycle hooks**: `on_train_start`, `on_train_batch_end`, `on_validation_end`, `on_save_checkpoint`, `on_exception`,...

fix: Use uv Python for MCore dataset compilation (#438)

9

## Description Changed Makefile to use `uv run python` instead of system `python3`, ensuring the compiled extension matches the uv Python environment. Also added `-undefined dynamic_lookup` linking flag for macOS...