TopFormer
TopFormer copied to clipboard
hustvl
TopFormer: Token Pyramid Transformer for Mobile Semantic Segmentation, CVPR2022
我用其他的训练模型转换后都能在手机上跑通，唯独SegFormer和Semantic FPN ConvMLP-S有问题。请问这两个模型的tnnproto文件或者在转换过程中有经过特殊处理吗，是否可以提供tnnproto文件。
C++推理模型结果全是1.4013e-45和0，这个如何解决？ 
Regardless of whether I use pre-training weights or not, the mIoU on the cityscapes dataset is only **55-59**, which is a far cry from the **70.7** in your paper, what...
To test Topformer's inference speed in mobile device, I followed tnn_runtime.md guide. But when I try `python3 tools/convert2onnx.py --input-img --shape 512 512 --checkpoint ` to convert model to ONNX, I...
运行命令：bash tools/dist_test.sh local_configs/topformer/topformer_base_512x512_160k_4x8_ade20k.py latest.pth 1（原本的sh要换成bash才能执行） 报错：AssertionError: Please specify at least one operation (save/eval/format/show the results / save the results) with the argument "--out", "--eval", "--format-only", "--show" or "--show-dir"
``` class CLSHead(nn.Module): def __init__(self, in_ch=384, out_ch=1000, dropout=0.2): super().__init__() self.pool = nn.AdaptiveAvgPool2d(1) self.bn = nn.BatchNorm2d(in_ch) self.dropout = nn.Dropout(p=dropout, inplace=True) self.l = nn.Linear(in_ch, out_ch) def forward(self, x): x = self.bn(x) x...
I have tried to pre-training the backbone+ppa+sase on imagenet, with convbnconv + linear as head. But I cannot reproduce the result of 75.3 on base model. So can you provide...
Hi, onnxsim is a wonderfull onnx graph optimizer. I think it may be useful for the deploy of topformer! It could remove some unuseful ops in onnx graph and sometimes...
