Created by: ddokupil

Cannot obtain any meaningful results when when trying to reproduce ResNet50 accuracy (76.63%/93.10%).

Using Image Net val dataset: val/659/ILSVRC2012_val_00046108.JPEG 659 val/659/ILSVRC2012_val_00044030.JPEG 659 val/659/ILSVRC2012_val_00042898.JPEG 659 val/659/ILSVRC2012_val_00041481.JPEG 659 val/659/ILSVRC2012_val_00039493.JPEG 659 val/659/ILSVRC2012_val_00037864.JPEG 659 val/659/ILSVRC2012_val_00037578.JPEG 659 val/659/ILSVRC2012_val_00033095.JPEG 659 val/659/ILSVRC2012_val_00031091.JPEG 659 val/659/ILSVRC2012_val_00030154.JPEG 659 val/659/ILSVRC2012_val_00029819.JPEG 659 val/659/ILSVRC2012_val_00029360.JPEG 659 val/659/ILSVRC2012_val_00028777.JPEG 659

And provided model I keep getting:

----------- Configuration Arguments ----------- class_dim: 1000 image_shape: 3,224,224 model: ResNet50 pretrained_model: /data/resnet_50/115 use_gpu: 0 with_mem_opt: 1

Test-0-score: [30.961004], class [550] Test-1-score: [16.557055], class [550] Test-2-score: [21.98035], class [550] Test-3-score: [14.857814], class [505] Test-4-score: [12.008922], class [550] Test-5-score: [19.638456], class [550] Test-6-score: [16.092129], class [813] Test-7-score: [24.678347], class [550] Test-8-score: [18.044142], class [550] Test-9-score: [22.066696], class [550] Test-10-score: [15.797653], class [550] Test-11-score: [20.66053], class [550] Test-12-score: [16.733633], class [482] Test-13-score: [15.197915], class [550] Test-14-score: [18.904078], class [550] Test-15-score: [27.324083], class [550] Test-16-score: [16.564373], class [550] Test-17-score: [14.2177515], class [813]

And:

----------- Configuration Arguments ----------- batch_size: 1 class_dim: 1000 image_shape: 3,224,224 model: ResNet50 pretrained_model: /data/resnet_50/115 use_gpu: 0 with_mem_opt: 1

Testbatch 0,loss 12.8082323074, acc1 0.0,acc5 0.0,time 0.33 sec Testbatch 10,loss 25.0121593475, acc1 0.0,acc5 0.0,time 0.16 sec Testbatch 20,loss 20.8600234985, acc1 0.0,acc5 0.0,time 0.16 sec Testbatch 30,loss 21.9461364746, acc1 0.0,acc5 0.0,time 0.16 sec Testbatch 40,loss 14.2183275223, acc1 0.0,acc5 0.0,time 0.17 sec Testbatch 50,loss 22.2879886627, acc1 0.0,acc5 0.0,time 0.19 sec Testbatch 60,loss 15.6957626343, acc1 0.0,acc5 0.0,time 0.16 sec Testbatch 70,loss 18.4210853577, acc1 0.0,acc5 0.0,time 0.16 sec Testbatch 80,loss 26.5067806244, acc1 0.0,acc5 0.0,time 0.17 sec

or

----------- Configuration Arguments ----------- batch_size: 32 class_dim: 1000 image_shape: 3,224,224 model: ResNet50 pretrained_model: /data/resnet_50/115 use_gpu: 0 with_mem_opt: 1

Testbatch 0,loss 16.9630355835, acc1 0.03125,acc5 0.03125,time 4.76 sec Testbatch 10,loss 16.6469154358, acc1 0.0,acc5 0.0,time 24.60 sec Testbatch 20,loss 16.7011985779, acc1 0.0,acc5 0.0,time 3.54 sec Testbatch 30,loss 13.4966154099, acc1 0.0,acc5 0.0,time 3.48 sec