Cedric Luo

icon location Shanghai, China icon location LLM, MLLM

Results 25 comments of Cedric Luo

Mask2Former training with V100 instead of A100 GPUs

There is env info in the beginning of log file, see https://download.openmmlab.com/mmdetection/v2.0/mask2former/mask2former_r50_lsj_8x2_50e_coco-panoptic/mask2former_r50_lsj_8x2_50e_coco-panoptic_20220326_224516.log.json

A question about comment in `mask2former_head`

@pixeli99 yes, you are right. I will update it later.

[Enhance] Add auto scale lr for maskformer_swin_l

@hhaAndroid All config files are added with auto_scale_lr which is disabled.

Fix swin backbone absolute pos_embed

please add a unit test in https://github.com/open-mmlab/mmdetection/blob/master/tests/test_models/test_backbones/test_swin.py for the modification

why still need to use conf_mask but not y_true[...,4]

> I m very confused with this, I still read know the loss_code but still don't know why, > what directly different with this, > > > ##### conf_mask =...

using jetson nano kit

Jetson nano kit can be used for model training (very lightweight model), but it is more suitable for model inference

import custom model error

ConvNeXt is not supported yet, #7281

import custom model error

Please check whether convnext exists on the sever and whether you use the expected the conda or pip environment.

import custom model error

In the picture, the path is `mmdet.models.backbones.csp_darknet.py`, not `mmdet.models.backbones._csp_darknet.py`

import custom model error

Please reinstall mmdet by `pip install -v -e .` in directory `mmdetection`.