Open DavidHuji opened 2 years ago
@DavidHuji You can just set the batch size to 1 with "TEST.IMS_PER_BATCH 1". Line 98 "rel_ind_i += offset" causes the problem. If you read the for loop, you will see how setting the batch size to 1 solves the problem.
Hi,
I get the following error when debugging with VG and RelDN:
IndexKernel.cu:93: operator(): block: [157,0,0], thread: [32,0,0] Assertion
index >= -sizes[i] && index < sizes[i] && "index out of bounds"
failed.The trigger is from scene_graph_benchmark/relation_head/reldn/reldn.py line number: 103 at: proposal_per_image.get_field("subj_box_features")[rel_ind_i[:, 0]] Actually the indexs at rel_ind_i are too big for proposal_per_image.get_field("subj_box_features").
It looks like something was not configured correctly in my env. I have downloaded the weights to the right path though.
Any idea?