search

jd730 / OICR-pytorch

Pytorch Implementation of Multiple Instance Detection Network with Online Instance Classifier Refinement
MIT License
75 stars 10 forks source link

The result of my mAP is very low #17

Closed fregulationn closed 4 years ago

fregulationn commented 4 years ago

In my training process, I use the run.sh as below:

CUDA_VISIBLE_DEVICES=2  python trainval_net_org.py --dataset pascal_voc --net vgg16 \
   --bs 2 --nw 8 --save_dir=results/test_single_9  \
   --lr 0.001 --cuda --disp_interval 100 --vis \
    --checkpoint_interval=5000 --model='oicr' --threshold=0.1

CUDA_VISIBLE_DEVICES=2 python test_oicr.py --dataset pascal_voc --net vgg16  --checkpoint 070000 --load_dir=results/test_single_9 --cuda --output_dir="test" --model="oicr" --vis

the GPU is TiTan V, pytroch version is 0.4.0 But I get the AP result like this : image I haven't found the reason yet.

jd730 commented 4 years ago

Hi @fregulationn could you tell me what trainval_net_org.py is? Is it same as original trainval_net.py?

Also did you run on the latest version?

(I am sorry that I mistakenly touched close issue button on my smartphone.)

fregulationn commented 4 years ago

the trainval_net_org.py is same as trainval_net.py I download the code by git clone git:://github.com/jd730/OICR-pytorch.git

thanks for your answer,my server is offline，so I can't have a try. I will try your suggestion in feature.