Caffe: Error Check failed: match_index[p] < gt_bboxes.size() when applied a pvanet as model in SSD

[Jiaqiww]

@weiliu89 Hello, I'm using a pvanet as the features extraction model in SSD. I have added test_loss layer and silence layer in my test net file. However, I got this as an error:

bbox_utils.cu: 595] Check failed: match_index[p] < gt_bboxes.size( ) (5 vs. 1)

What does this error mean? Thank you in advance!

[Jiaqiww]

Could any one tell me what do they mean, match_index and gt_bboxes.size()?

[weiliu89]

Check the code.

[wantai2017]

@Jiaqiww @liyuanyaun@louis0802 Hello, I met the same problem: Check failed: match_index[p] < gt_bboxes.size() (4vs1) when I added another data layer and test loss, could anybody tell me how to fix it? Thanks a lot!

[wujiankepj]

@wantai2017 I met the same problem. I add loss_weight=1 to the final layer of the new net, and this problem solved. The reason is the new net do not have a loss layer if we add other layer with multi_box_loss_layer. This wil trigger a bug (may be) in bbox_util.cpp

[angiend]

@wujiankepj

I met the same problem. I add loss_weight=1 to the final layer of the new net, and this problem solved. The reason is the new net do not have a loss layer if we add other layer with multi_box_loss_layer. This wil trigger a bug (may be) in bbox_util.cpp

would you explain the solution more ?thanks,
I run Mobile-v1-SSD,and met the same problem,but i donnot kown how to solver it,thanks again!