search

Hwang64 / MLKP

CVPR18 Paper: Multi-scale Location-aware Kernel Representation for Object Detection
107 stars 24 forks source link

fail to reproduce the AP with the model downloaded from BaiduYun #9

Open UeFan opened 6 years ago

UeFan commented 6 years ago

After successfully make the caffe-mlkp/ and lib/ , I download the ResNet model provided in BaiduYun. Run the test_net.py for voc_2007_test, but only get Mean AP = 0.7224.

I also tried to train the VOC2007, with a imagenet-pretrained-VCC16 model (~500MB), but after 50000 iters, I get an AP around 0, I think there must be something wrong with my implementation. Could you give me some clue?

For training the voc2007, is there anything need to be modified in config.py, solver.prototext or somewhere else?

This is the picture i took during training.

image
qilei123 commented 6 years ago

UP, I meet the same problem! Have you got the solution?