chengyangfu / caffe

Caffe: a fast open framework for deep learning.
http://caffe.berkeleyvision.org/
Other
169 stars 47 forks source link

The accuracy of your SSD_513_COCO(model weights) is 0.13% #26

Open youngwanLEE opened 6 years ago

youngwanLEE commented 6 years ago

I downloaded your model weight and tested it on COCO-minival as follow:

./build/tools/caffe train \
--solver="models/ResNet-101/coco/SSD_513x513/test_solver.prototxt" \
--snapshot="models/ResNet-101/coco/SSD_513x513/ResNet-101_coco_SSD_513x513_iter_250000.solverstate" \
--gpu 0,1,2,3 2>&1 | tee jobs/ResNet-101/coco/SSD_513x513/ResNet-101_coco_SSD_513x513.log

the result is " Test net output #0: detection_eval = 0.00131212 "

What's the problem?

I guess that many people faced this problem.

gk19950815 commented 4 years ago

I have this problem, too.Have you solved it?