wangle53 / transcd Goto Github PK

In offered train.txt, the name of label is mask**.png, be the name of label in original VL_CMU_CD is gt**.png, so how to generate maskGT? can I just change their name?

The metrics of every category are all the same in test_score.txt.
My arguments:
--net_cfg SViT_E1_D1_32 --train_cfg CDNet_2014 --save_changemap True

I cropped some below:
Category:badWeather,precision:0.9654355468223322,oa:0.8092331883428242,recall:0.1367026792675585,f1:0.23949377961693205,kappa:0.19565049499554038
Category:baseline,precision:0.9654355468223322,oa:0.8092331883428242,recall:0.1367026792675585,f1:0.23949377961693205,kappa:0.19565049499554038
Category:cameraJitter,precision:0.9654355468223322,oa:0.8092331883428242,recall:0.1367026792675585,f1:0.23949377961693205,kappa:0.19565049499554038
Category:dynamicBackground,precision:0.9654355468223322,oa:0.8092331883428242,recall:0.1367026792675585,f1:0.23949377961693205,kappa:0.19565049499554038
Category:intermittentObjectMotion,precision:0.9654355468223322,oa:0.8092331883428242,recall:0.1367026792675585,f1:0.23949377961693205,kappa: 0.19565049499554038

Do you consider transferring the model to medical image recognition or segment? Medical images always include multiple targets or structures, especially in longitudinal monitoring in vivo, can your model be applied to track or swiftly extract different structures from these images?

Lucas

Can you please give some descriptions about the 6 different models?
SViT_E1_D1_16, SViT_E1_D1_32, SViT_E4_D4_16, SViT_E4_D4_32, Res_SViT_E1_D1_16 and Res_SViT_E4_D4_16.

How are they different? And which one is faster and more accurate, etc?

using arguments: '--net_cfg SViT_E1_D1_32 --train_cfg CDNet_2014 --save_changemap True'
test.py returns "No such file or directory: './CDNet_2014/data/test.txt' "

hi, What is the difference between macro eva and micro eva in test.txt?

CDNet-2014 dataset format is not the same as train.py reading format which is based on t1, t2 and gt.
Is there any other resources for CDNet-2014 dataset which is match to your code?

Interesting work. How can I test the pretrained model on a single image pair instead of testing a whole directory?

Basically given two images I want to get the changed parts.

Hi, using transformer was very interesting. Are you aware of any newer work or more advanced models that might work better than transCD? Done by you or others.

您好，想请教一下什么情况下F1值和其余都为0，但是oa会九十多分，我是在您TransCD代码上增加了一个模块，关于计算指标的部分我没有进行变动

import _ssl ImportError: DLL load failed: 找不到指定的模块。

The URL (https://ghsi.github.io/proj/RSS2016.html) seems down.

May I download the VL-CMU-CD dataset somewhere else?
Thanks.

Hi -,

Can you please kindly share the CDNet-2014 data because the changedetection.net website is not working anymore?

Thank you!!

thanks for sharing your great work.when i use this code train my model like vl_cmu_cd format,but it did't work,the error is the following:
return _VF.broadcast_tensors(tensors) # type: ignore[attr-defined]
RuntimeError: The size of tensor a (512) must match the size of tensor b (3) at non-singleton dimension 4
look forward for your reply

Thank you for sharing the code.

I have a question I'd like to answer from you.

At make_dataset.py, line 45, gt adds a dimension, which causes The error"RuntimeError: The size of tensor a (512) must match The size of tensor B (3) at non-singleton Dimension 4". What are the considerations for this step?

Best regards

VL-CMU-CD can not be downloaded

After the training is complete, there is no change map in the savechangemap folder