xinge008 / scda Goto Github PK

Hi there,

I'm trying to use my own dataset with this code. I've encountered the following error message and tried to fixed it, but after a couple of attempts I gave up.

It would be great if somebody can give me a hint.

The error message:

RuntimeError: Given groups=1, weight of size [64, 3, 3, 3], expected input[1, 1, 1280, 1280] to have 3 channels, but got 1 channels instead

Maybe some useful information:

The pictures I've used were originally in JPEG Format and were converted with mogrify to PNG8.

Code is run in docker container with the following image:
docker.io/pytorch/pytorch:0.4.1-cuda9-cudnn7-devel

Tested on two independent host systems, RTX2070 Super and GTX 1060. Same error on both machines.

Thank you in advance.

The result of the original RPN will be based on a threshold, such as 0.5 to determine which boxes are left. But for an image without ground truth, how to determine which frames are useful?

Hi, is it possible to train the model on Google Colab?
If yes, can you please brief the steps that are needed to be followed?

I input common:
~/code/SCDA-master$ cd datasets/pycocotools & make
and error.
Could you tell me how to process it clearly?
Thank you!

Same as the title, while I read your code, i find a question:
dis_model / dec_model / dis_model_patch which is generator/discriminator/weighting estimator??
BTW, the code of update process is not vague for me. (From line 564 in faster_rcnn_train_val.py)
Thank you for your answering!

I found that in the foggy_val.txt you provided in the README contains 1476 images. But the results in [4] are reported on 500 val images. And your baseline is also much higher than [4], whereas the only difference is the val set. Does that means your improvement only comes from this slight change?

I want to just use one image to show the result, how to do it?

Hi, I am wondering how many gpus did you use in all the exps?