Inference with single camera image

[hocherie]

Hi, thanks for the fantastic work.

I was able to run your segmentation pipeline with 6-images with your inference script and pretrained weights.

However, when I run with just the front image, the results look very strange. (left) ground truth, (right) result

In order to just use a single image, I did data['img'][0].data[0] = data['img'][0].data[0][:,0,:,:,:] same for img_metas.

Did we miss a step or is this the expected performance? Should we retrain? Thanks!

[wadefrank]

@hocherie Hi, can I ask you how to inference with a single camera image? Which part of the code needs to be modified? Can you give me an example? Thanks a lot!