Unexpected results when using weights downloaded from the dropbox link.

[SophiaSarah]

Using weights from Dropbox link, I got unexpected results when deblurring images from GoPro and YOLO dataset as same as ones in the ./DeblurGAN/images/. Is there anything wrong? @:~/DeblurGAN$ python test.py --dataroot ./images/test/ --model test --dataset_mode single --learn_residual --resize_or_crop no_change ------------ Options ------------- aspect_ratio: 1.0 batchSize: 1 checkpoints_dir: ./checkpoints dataroot: ./images/test/ dataset_mode: single display_id: 1 display_port: 8097 display_single_pane_ncols: 0 display_winsize: 256 fineSize: 256 gan_type: wgan-gp gpu_ids: [0] how_many: 5000 input_nc: 3 isTrain: False learn_residual: True loadSizeX: 640 loadSizeY: 360 max_dataset_size: inf model: test nThreads: 2 n_layers_D: 3 name: experiment_name ndf: 64 ngf: 64 no_dropout: False no_flip: False norm: instance ntest: inf output_nc: 3 phase: test resize_or_crop: no_change results_dir: ./results/ serial_batches: False which_direction: AtoB which_epoch: latest which_model_netD: basic which_model_netG: resnet_9blocks -------------- End ---------------- CustomDatasetDataLoader dataset [SingleImageDataset] was created ---------- Networks initialized ------------- ……

model [TestModel] was created [2018-06-20 16:05:32,425] INFO: Socket refused connection, running socketless process image... ['./images/test/test1_b.jpg'] process image... ['./images/test/yolo_b.jpg'] ————————————-———————————————————————— My results are as followed: test1_b.jpg

test1_b_fake_B.png

The corresponding result as you've shown:

yolo_b.jpg

yolo_b_fake_B.png

The corresponding result as you've shown:

[KupynOrest]

Hello, I've compressed & changed those images to put into paper, so you may not receive good results on them, please try to run the model on original images