search

hector6298 / mcmt_gnn

Multi-Object Multi-Camera tracking using graph neural networks. Under active development. My MSc project.
GNU General Public License v3.0
0 stars 1 forks source link

Add detection filter for SC tracks on evaluation #18

Closed hector6298 closed 1 year ago

hector6298 commented 1 year ago

Why is this filtering entire ids?

        for i in unique_ids:

            frame_min = np.min(mtsc[mtsc['id'].values == i]['frame'].values)
            frame_max = np.max(mtsc[mtsc['id'].values == i]['frame'].values)
            mtsc['ybot'] = (mtsc['ymin'].values + mtsc['height'].values)
            mtsc['xbot'] = (mtsc['xmin'].values + (mtsc['width'].values / 2)).astype(int)
            initial_pos = np.asarray(
                [mtsc[(mtsc['id'].values == i) & (mtsc['frame'].values == frame_min)]['ybot'].values[0],
                 mtsc[(mtsc['id'].values == i) & (mtsc['frame'].values == frame_min)]['xbot'].values[0]])
            final_pos = np.asarray(
                [mtsc[(mtsc['id'].values == i) & (mtsc['frame'].values == frame_max)]['ybot'].values[0],
                 mtsc[(mtsc['id'].values == i) & (mtsc['frame'].values == frame_max)]['xbot'].values[0]])

            initial_roi = roi[int(initial_pos[0]), int(initial_pos[1])]
            final_roi = roi[int(final_pos[0]), int(final_pos[1])]

            if initial_roi == 0 or final_roi == 0:
                mtsc = mtsc[mtsc['id'] != i]
hector6298 commented 1 year ago

How is the provided ground truth file generated?

hector6298 commented 1 year ago

check if there are entire ids being eliminated by this script