TFFRCNN VGG training example seems learned nothing

VGG training example seems learned nothing

Open Asiapenolove opened this issue 7 years ago • 4 comments

Hi, I have training VGG - faster-rcnn example follows the Readme.md

with

python ./faster_rcnn/train_net.py --gpu 0 \
    --weights ./data/pretrain_model/VGG_imagenet.npy \
    --imdb voc_2007_trainval --iters 70000 \
    --cfg  ./experiments/cfgs/faster_rcnn_end2end.yml \
    --network VGGnet_train --set EXP_DIR exp_dir

and tested with

python ./faster_rcnn/test_net.py --gpu 0 --weights ./output/exp_dir/voc_2007_trainval/VGGnet_fast_rcnn_iter_70000.ckpt --imdb voc_2007_test --cfg ./experiments/cfgs/faster_rcnn_end2end.yml --network VGGnet_test

AP for aeroplane = -1.0000 AP for bicycle = -1.0000 AP for bird = -1.0000 AP for boat = -1.0000 AP for bottle = -1.0000 AP for bus = -1.0000 AP for car = -1.0000 AP for cat = -1.0000 AP for chair = -1.0000 AP for cow = -1.0000 AP for diningtable = -1.0000 AP for dog = -1.0000 AP for horse = -1.0000 AP for motorbike = -1.0000 AP for person = 0.0183 AP for pottedplant = -1.0000 AP for sheep = -1.0000 AP for sofa = -1.0000 AP for train = -1.0000 AP for tvmonitor = -1.0000 Mean AP = -0.9491

I have edited the test.py to skip model exist checking (since tf r1.1 saving as .index .meta .data three file instead one) But I didn't think that is the reason fail training.

Are there anyone with this similar problem??

my tensorflow version r1.1