search

weiliu89 / caffe

Caffe: a fast open framework for deep learning.
http://caffe.berkeleyvision.org/
Other
4.77k stars 1.67k forks source link

problem in detection using fc reduced VGGNet #182

Open sunejas opened 7 years ago

sunejas commented 7 years ago

Hi, I am trying to run detection on an image using the fc_reduced VGG net from https://gist.github.com/weiliu89/2ed6e13bfd5b57cf81d6

The command i use is: .build_release/examples/ssd/ssd_detect.bin -confidence_threshold 0.6 -out_file outfile.txt models/VGGNet/VGG_ILSVRC_16_layers_fc_reduced_deploy.prototxt models/VGGNet/VGG_ILSVRC_16_layers_fc_reduced.caffemodel list_file

No detection happens in this case and the outfile is empty.

Although when I use the SSD300 model, the object detection happens correctly and outfile is populated with correct detections. I used similar command as above: .build_release/examples/ssd/ssd_detect.bin -confidence_threshold 0.6 -out_file outfile.txt models/VGGNet/VOC0712/SSD_300x300/deploy.prototxt models/VGGNet/VOC0712/SSD_300x300/VGG_VOC0712_SSD_300x300_iter_60000.caffemodel list_file

Could anyone please guide me what could be the reason fc_reduced model detection is not working ?

weiliu89 commented 7 years ago

That is a classification model (VGGNet with subsampled fc6/fc7 parameters)