Inference colab file not running

[Szransh]

Inference colab file not running

[Syzygianinfern0]

As far as I remember, the codebase is heavily sensitive to CUDA/CUDnn versions. Perhaps Colab has updated their versions and our code does not run anymore on it :(

[Szransh]

Can you please help us by suggesting how we should run the same on colab? or providing us with the correct versions of CUDA/CUDnn versions so maybe I can try running it locally.

[Syzygianinfern0]

• The model code might need some updating to make it work with colab so that it supports newer CUDA versions. This might need some digging into each of these functions which say they do not exist anymore and find their new equivalent APIs.
• You can find the versions with which we ran our experiments here. You can try setting this up locally and you should have it up and running!

[Szransh]

Hi, there is an inherent mismatch between the torch and torch vision modules you gave. I could run the script using Pytorch 1.8.0 with torch vision 0.9.0 and Cuda toolkit 11.1 and download the detectron2 version corresponding to the same while utilising the requirements.txt file for other packages. The image is segmented for now.
However, on running the following command
input = cv2.imread("./images/test.jpg") # recieving the input image
input = input[:,:,::-1]
showoutput(input)

I get the error: No such file or directory: './outputs/data.json
please help me with this

[Szransh]

No such file is generated. Is there any fix?? I need the json files so that I can have the coordinates of bounding box which I can later use for cropping.

[Syzygianinfern0]

Where is this showoutput(input)?

[Szransh]

It is part of your colab notebook. In your notebook itself the same error is showing as uploaded by you.

[Szransh]

Is there any way that I can access the json file containing the co-ordinates for each line because I wanted use that information for further processing.

[Syzygianinfern0]

I see, thanks for bring this to my attention. I just realized that I am infact not the author of that colab notebook. From the
revision history, it is from my co-authors and other collaborators at the lab.

But from looking at code from colab notebook, the JSON seems to be from performing manual annotations using an annotation tool. I am not totally sure. I suppose you can find GT annotations from the dataset.

There is also an older version of the colab notebook in its revision history which I located and does not have this error: Link.

I shall get back to you on this soon!

[Syzygianinfern0]

I would recommend you to use Seamformer instead which is a follow-up work from our lab.