search

Zhangjinso / PISE

123 stars 28 forks source link

generating samples by the pretrained model #34

Closed Mathilda88 closed 5 months ago

Mathilda88 commented 3 years ago

Hello,

Thanks for sharing the great work. I'm trying to generate the samples using the pretrained model. But unfortunately my results are so dim, something like this: Mine: fashionMENJackets_Vestsid0000488201_7additional_fashionMENJackets_Vestsid0000488201_1front_vis Previously available by the authors: fashionMENJackets_Vestsid0000488201_7additional_fashionMENJackets_Vestsid0000488201_1front_vis

Any help is greatly appreciated by the entire community.

Zhangjinso commented 3 years ago

Hi, it is strange to get this kind of outputs. Could you give more information about your procedures for this output? Like the platform you used and the outputs after running...

Mathilda88 commented 3 years ago

Thanks for your response. Actually, I'm on Fedora Linux, I tried different types of GPUs on a cluster, RTX8000 and V100 but the results are quite the same, with an old_size (256,256) and all the default setting of parameters. Additionally, I'm using the following packages:

python=3.6, pytorch=1.2, cudatoolkit=10.0 torchvision, anaconda 0.0.1.1 bleach 1.5.0 block-extractor-cuda 0.0.0 certifi 2021.5.30 cffi 1.14.6 cycler 0.10.0 dataclasses 0.8 decorator 4.4.2 dominate 2.6.0 html5lib 0.9999999 imageio 2.9.0 importlib-metadata 3.4.0 kiwisolver 1.3.1 local-attn-reshape-cuda 0.0.0 matplotlib 3.3.4 mkl-fft 1.3.0 mkl-random 1.1.1 mkl-service 2.3.0 natsort 7.1.1 networkx 2.5.1 numpy 1.19.2 olefile 0.46 opencv-python 4.5.3.56 pandas 1.1.5 Pillow 8.3.1 pip 21.0.1 pycparser 2.20 pyparsing 2.4.7 python-dateutil 2.8.2 pytz 2021.1 PyWavelets 1.1.1 resample2d-cuda 0.0.0 scikit-image 0.17.2 scipy 1.5.4 setuptools 58.0.4 six 1.16.0 tensorflow-tensorboard 0.4.0 tifffile 2020.9.3 torch 1.2.0 torchvision 0.4.0a0 tqdm 4.62.2 typing-extensions 3.10.0.0 wheel 0.36.2 zipp 3.4.1

Mathilda88 commented 3 years ago

Here are some examples: fashionWOMENBlouses_Shirtsid0000546609_3back_fashionWOMENBlouses_Shirtsid0000546609_2side_vis fashionWOMENBlouses_Shirtsid0000750504_7additional_fashionWOMENBlouses_Shirtsid0000750504_2side_vis fashionWOMENBlouses_Shirtsid0000787301_1front_fashionWOMENBlouses_Shirtsid0000787301_4full_vis fashionWOMENCardigansid0000011903_1front_fashionWOMENCardigansid0000011903_2side_vis

Zhangjinso commented 3 years ago

Hi, I just test the current code with the checkpoints (with titan x on ubuntu), and the results are ok. I have never been seen this kind of result before. Sorry

Mathilda88 commented 3 years ago

Thanks for your response. May you keep the topic open to get the chance of leveraging from other experiences.

pilgrim00 commented 3 years ago

Hello, I would like to ask that whether the FID Scores is tested by the test_result you provided ?

Zhangjinso commented 3 years ago

Yes. The fid is computed as 256X256 between the generated images and the training images using this.