Using pascal-multi with images from the MOT challenge videos

I know that pascal-multi hasn’t actually been discussed yet in a lesson but I just couldn’t help myself and had to play around with it over the last days.

I am trying to load custom images into the notebook, which are images taken from the MOT (Multiple Object Tracking) challenge videos

https://motchallenge.net/

I think I got it kind of work but the grid won’t align with the image, any hints?

I simply added the cells in the screenshot below the first image in the Testing section of the pascal-multi notebook.

Here is the content of the cells again as text if you want to try it yourself:

tfms = tfms_from_model(f_model, sz, crop_type=CropType.NO)
t = tfms[0]
im = t(open_image(“mot.jpg”))
p = learn.predict_array(im[None])
x,y = V(x),V(y)
bbox,clas = get_y(y[0][0], y[1][0])
i2 = open_image(“mot.jpg”)
fig, ax = plt.subplots(figsize=(7,7))
torch_gt(ax, i2, bbox, clas)

mot.jpg:

4 Likes

You need to plot x[0], which is after transforms, not im, which is before transforms.

1 Like

I am working for the same task. Any codes to share? Thanks