gpu/shrinking.cu error when handling large images

Issue #58 open
Anonymous created an issue

We are trying to train our model, but facing the following error.

2014-08-22 12:39:35 {7f70ce3fea00} [ SoftCascadeOverIntegralChannelsModel ] : Will ignore the trivial cascade thresholds of the detector model Using score_threshold == 0.00000 2014-08-22 12:39:35 {7f70ce3fea00} [ SoftCascadeOverIntegralChannelsModel ] : Will ignore the trivial cascade thresholds of the detector model

0% 10 20 30 40 50 60 70 80 90 100% |----|----|----|----|----|----|----|----|----|----| ****~~~~ /src/applications/bootstrapping_lib/../../../src/objects_detection/integral_channels/gpu/shrinking.cu:215 (void doppia::integral_channels::bind_input_channels_texture(const gpu_channels_t&)): invalid argument Aborted (core dumped)

Can you let me know how we can fix it? Does this error come from the properties of our data? (aspect ratio, image size, number of images, ...)

Comments (8)

  1. Rodrigo Benenson repo owner

    We have observed this error when using (very) large images (the exact size depends on the capacity of the specific GPU and CUDA version).

    If you must feed large images, we have a fix for this (at the price of slightly lower speed). Let me know if you need it and I will add it to v2.

  2. Rodrigo Benenson repo owner

    Was the previous answer what you were looking for ?

    As mentioned if you really need to detect on large image, re-open the issue and I will look into importing the patch.

    For general use I guess most people care about fast detection on normal size image, and being able to handle large images.

    Closing the issue until further notice.

  3. money618

    Hi, I am trying to test the headhunter model in Ubuntu 13.10 with CUDA6.0. When I detect on large image, I met the same error as above. The largest size of an input image I can obtain correct result is 700*700. Would you tell me I can fix this error? Thanks very much!

  4. liuwei

    hello´╝îmoney618, I am also trying to test the headhunter model,When I detect on large image, I met the same error as above.Would you tell me I can fix this error? Thanks .

  5. Log in to comment