Finetuning error: "Creating layer conv14_2_mbox_conf_flat" --> "Check failed: std::equal(top_shape.begin() + 1, top_shape.begin() + 4, shape.begin() + 1) "

[gombru]

Hello,

I'm trying to fine-tune MobileNet-SSD using a modified version of COCO. I have created the lmdb's of my modified COCO version using SSD. Then I have used gen_model.sh 21 to generate the net prototxt. When I call train.sh to start eh finetuning, I get this error:

0209 12:37:34.576712 15523 layer_factory.hpp:77] Creating layer conv14_2_mbox_conf_flat F0209 12:37:34.581527 15545 annotated_data_layer.cpp:205] Check failed: std::equal(top_shape.begin() + 1, top_shape.begin() + 4, shape.begin() + 1)

Any idea about the reason? I'm not modifying the net architecture, so its weird. Thanks!

[zhangnn016]

I got this too, has any one solved this?

[apatti]

I am facing same issue, any solutions?

[stalagmite7]

any solution to this issue yet? I just encountered this

[urbaneman]

I had this issue too? Did you solution this? Maybe, the lmdbs have error? But I'm not sure. I used my old lmdbs in the same Net and parameters, the issue did not appear.

[shiva13425]

Hi, have anyone solved the issue? Kindly share your comments.

[MintHacker]

I solved it. Firslty, you should check the image size. Is there any image less than 300 pixels in width or height? If so, you need to sort them out. If there is no, may the RGB image is recognized as Gray, you can add "force_color: true" to the data param, both train.prototxt and test.prototxt need to be modified. Then, it can be solved.