F1103 20:53:52.339550 156870 io.cpp:63] Check failed: fd != -1 (-1 vs. -1) File not found: ../models/VOC300TRAINED_4e04e3.caffemodel about des

[bairdzhang]

Hi, please download the pre-trained model https://github.com/bairdzhang/des/blob/des/initial_models/VOC300TRAINED_4e04e3.caffemodel and change the --weights= option in DES300x300.sh to specify that pre-trained model.

[taomrzhang]

Hi, please download the pre-trained model https://github.com/bairdzhang/des/blob/des/initial_models/VOC300TRAINED_4e04e3.caffemodel and change the --weights= option in DES300x300.sh to specify that pre-trained model.

I am sorry I can not solve it. when I put 'VOC300TRAINED_4e04e3.caffemodel' in ../models or ../models/VGGNet, I met the same error again. and in DES300*300.sh the --weights="../models/VOC300TRAINED_4e04e3.caffemodel" \
thank you

[bairdzhang]

In jobs/VGGNet/VOC0712/DES300x300/DES300x300.sh, it first enter the root folder of this repo (/workspace/release/caffe-ssd as the example shell script), and run ./build/tools/caffe with the pre-trained model ../models/VOC300TRAINED_4e04e3.caffemodel.
I guess the path (../models/VOC300TRAINED_4e04e3.caffemodel) is based on the current work folder, which is /workspace/release/caffe-ssd, so the pre-trained model should be located in /workspace/release/model/VOC300TRAINED_4e04e3.caffemodel.

A safe way is to try the absolute path of the pre-trained model in the shell script.

[taomrzhang]

In jobs/VGGNet/VOC0712/DES300x300/DES300x300.sh, it first enter the root folder of this repo (/workspace/release/caffe-ssd as the example shell script), and run ./build/tools/caffe with the pre-trained model ../models/VOC300TRAINED_4e04e3.caffemodel.
I guess the path (../models/VOC300TRAINED_4e04e3.caffemodel) is based on the current work folder, which is /workspace/release/caffe-ssd, so the pre-trained model should be located in /workspace/release/model/VOC300TRAINED_4e04e3.caffemodel.

A safe way is to try the absolute path of the pre-trained model in the shell script.

thank you very much. I have solved it in that way.