-
> Traceback (most recent call last):
> File "/home/pan/zhaoda_code_important/zhaoda_code/nncf-develop/examples/torch/semantic_segmentation/main.py", line 34, in
> from examples.torch.semantic…
-
when I use 'python3 train.py' as readme.
__error :__
Downloading /home/abc/.torch/models/resnet50-25c4b509.zip from https://hangzh.s3.amazonaws.com/encoding/models/resnet50-25c4b509.zip...
Trace…
-
你好,请教一下单机多卡问题,
不管设置gpu数量为多少,实际只使用了一张卡
默认的bs=8,单卡显存可以吃满,改为16还是一张卡,也没有溢出
无任何报错
-
"icnet_resnet50_197_0.710_best_model.pth" is not exist?
-
The network I used is **ICnet_resnet_r50_d8**. I want to change the maxpool layer of backbone for my project. The goal is to change the maxpool's kernel size from 3 to 2 and padding from 1 to 0.
The …
-
请问ICNET是完全按照论文中的进行的么,原图中1/16,1/8的预测结果都有加入到下一次CCF中,1/4的预测也没有看到存在位置。另一个问题是,1/4的预测结果是由1/16/1/8的预测结果再次上采样得到的的么 ? 在代码中没有体现
-
看到icnet的多路输出,想在其他模型中使用,有一下问题:
1.最后的损失是加和,那反向传播的时候也会多路反向传播,然后梯度叠加么?
2.多路损失权重设置一般是什么样的,看icnet是【1.0,0.4,0.16】
3.尝试在改进的unet上使用 ["lovasz_softmax_loss","softmax_loss"] 比重是0.2 0.8,但是没有效果上的提升,请问这样的原因是不是数据集…
-
**Submitting author:** @georgeyiasemis (George Yiasemis)
**Repository:** https://github.com/NKI-AI/direct
**Branch with paper.md** (empty if default branch): joss-submission
**Version:** v1.0.1
**…
-
I'm trying to run whole evaluation command inside `train_icnet.sh`
the command is:
```
python val.py --data_set cityscapes \
--data_dir "/nas/dataset/CityScapes" \
--data_list "./data/cityscapes/…
-
语义分割领域常用auxiliary loss来优化训练,官方提供的pdseg文件loss.py中定义了multi_softmax_with_loss()函数,并且在config.py文件中也定义了cfg.MODEL.MULTI_LOSS_WEIGHT。如果要实现auxiliary loss,是否只要在相应的模型文件中返回logits的元组,并且在config.py文件中设置logits的元组所对应…