Created by: sfraczek

Hi, Resnet50 eval result is wierd on pre-trained weights from https://github.com/PaddlePaddle/models/tree/develop/fluid/image_classification. I have downloaded develop branch for both Paddle and models

When I try eval:

CPU_NUM=1 FLAGS_use_mkldnn=0 python eval.py \                
--use_gpu False \
--batch_size 32 \
--model ResNet50 \
--pretrained_model models/resnet_50/115

I get

Testbatch 0,loss 17.3628101349, acc1 0.03125,acc5 0.03125,time 11.99 sec
Testbatch 10,loss 16.5818500519, acc1 0.0,acc5 0.0,time 10.16 sec
Testbatch 20,loss 16.7979774475, acc1 0.0,acc5 0.0,time 10.11 sec