lsnet
lsnet copied to clipboard
LSNet: See Large, Focus Small [CVPR 2025]
缺陷
目前,开源代码与论文所述的网络结构不完全一致,尤其以LS Block最为严重
关于环境配置问题
目前版本为torch2.4.1,cuda121,mmcv-full=1.7.2均按照readme配置,且对mmcv内部代码修改完毕,总是报错:ImportError: /opt/conda/envs/MC/lib/python3.8/site-packages/mmcv/_ext.cpython-38-x86_64-linux-gnu.so: undefined symbol: _ZN2at4_ops10zeros_like4callERKNS_6TensorEN3c108optionalINS5_10ScalarTypeEEENS6_INS5_6LayoutEEENS6_INS5_6DeviceEEENS6_IbEENS6_INS5_12MemoryFormatEEE。显示为当前mmcv有些扩展有些没法用高级的torch版本编译,请问具体配置应该是怎样的?
怎么使用自己的数据集进行训练,配置哪个文件设置标签和数据集,启动训练是用main.py 还是 detection 里面的train.py 。怎么设置GUP的哪个卡
您好,在代码中,我发现通过LKP生成的权重形状为(B, C//G, ks*ks, H, W),即model/lsnet.py中的第195行,这说明论文中wi的形状为(C//G, ks, ks),但在文中的介绍如下:  这里似乎有些矛盾,我不确定我理解的对不对,请您指教
看到代码中有cuda 底层编程,想请问下该模型是否可以导出为onnx呢
import timm model_list = timm.list_models() print(model_list) The above is my code. The following error was reported after execution: Traceback (most recent call last): File "d:\Programmer\important things\lsnet-master\fy-test.py", line 4, in model...
Many thanks for your excellent work and for sharing it with the community. I noticed that the SKA module includes custom forward and backward functions. Please correct me if I’m...
The system report error "cannot import name 'custom_fwd' from 'torch.amp‘" due to the call "from torch.amp import custom_fwd, custom_bwd" in ska.py, so,how to fix thie error?
但是lsnet的训练又是基于224的我不知道是否可行。还是说会有问题。