SegFormer

SegFormer copied to clipboard

How to convert the model to tensorrt or onnx?

2

For robot implementation, we need onnx or openvino version of segformer, but currently, I found that segformer is not to to be converted to those versions, Does anyone can help...

yuchenlichuck

How can Fig. 6 be reproduced?? (Effective Receptive Field)

1

Loparch

checkpoint: segformer.b5.1024x1024.city.160k.pth CS-val: +--------+-------+-------+-------+ | Scope | mIoU | mAcc | aAcc | +--------+-------+-------+-------+ | global | 82.39 | 89.24 | 96.73 | +--------+-------+-------+-------+ CS-test: +--------+------+------+------+ | Scope | mIoU...

cht619

What is the image size used to calculate the flops for each of the three datasets? For Cityscapes it looks like the image size is 2048x1024, what about the other...

wzhiyan

ADE B3 model SS mIoU doesn't match paper Table 1.

3

Run `python tools/test.py local_configs/segformer/B3/segformer.b3.512x512.ade.160k.py segformer.b3.512x512.ade.160k.pth` with given pretrained checkpoint in this repo. And got ``` +--------+-------+-------+-------+ | Scope | mIoU | mAcc | aAcc | +--------+-------+-------+-------+ | global | 45.48...

gathierry

The first program, I think there is a connection between the top and bottom. `import torch import torch.nn as nn import pytorch_lightning as pl class LightningSegformerForSemanticSegmentation(pl.LightningModule): def __init__(self, segformer): super().__init__()...

Wang-taoshuo

ModuleNotFoundError: No module named 'mmcv._ext'

3

I installed mmcv version 1.3.0 based on the instructions of the repo but it keeps raising an error "ModuleNotFoundError: No module named 'mmcv._ext'". Which version of mmcv should I use?

bekhzod-olimov

why does the model have to pre-trained on ImageNet-1k datasets? Did you test results directly on ADE20K without using pre-training?

1

754467737

# Welcome update to OpenMMLab 2.0 I am Vansin, the technical operator of OpenMMLab. In September of last year, we announced the release of OpenMMLab 2.0 at the World Artificial...

vansin

Solving #73 by adding .contiguous() after transpose and permutation.

Jayden9912