aelnouby / Text-to-Image-Synthesis

Pytorch implementation of Generative Adversarial Text-to-Image Synthesis paper
GNU General Public License v3.0
405 stars 89 forks source link

Mapping to dataset folders. #26

Open idigitopia opened 5 years ago

idigitopia commented 5 years ago

Can someone please point out where to map these folders?

flowers_images_path: '/export/mlrg/aelnouby/projects/GANs/flowers_dataset/' flowers_embedding_path: '/export/mlrg/aelnouby/projects/GANs/flowers_dataset/flowers_icml/' flowers_text_path: '/export/mlrg/aelnouby/projects/GANs/flowers_dataset/cvpr2016_flowers/text_c10/' flowers_dataset_path: '/scratch/aelnouby/text2image/flowers.hdf5'

flowers_val_split_path: '/export/mlrg/aelnouby/projects/GANs/flowers_dataset/flowers_icml/valclasses.txt' flowers_train_split_path: '/export/mlrg/aelnouby/projects/GANs/flowers_dataset/flowers_icml/trainclasses.txt' flowers_test_split_path: '/export/mlrg/aelnouby/projects/GANs/flowers_dataset/flowers_icml/testclasses.txt'

I have downloaded the cvpr2016_flowers dataset and have found the text_c10 folder, it also consisted of valClasses.txt trainclasses.txt and testclasses.txt

I am not too sure where to point these two particular paths: flowers_images_path: '/export/mlrg/aelnouby/projects/GANs/flowers_dataset/' flowers_embedding_path: '/export/mlrg/aelnouby/projects/GANs/flowers_dataset/flowers_icml/'

shahhaard47 commented 4 years ago

I think two source repositories are important. Get the cvpr2016 stuff from https://github.com/reedscot/cvpr2016 and icml2016 stuff from reedscot/icml2016.

To answer your question:

Side note : it seems the default parameters for *_split_path variables links to flowers_icml/*.txt.

Lastly, if you use the provided .hd5 file in the Datasets section, you don't need any of the other files. You only need all the data if you want to convert the data into hd5 file yourself. If you're not able to use the provided hd5 file for version incompatibility issues, try to add this requirements.txt file and install the appropriate versions with pip install -r requirements.

mahinash26 commented 4 years ago

If we are using hd5 file, then how will we use that?Kindly guide me plz.

I think two source repositories are important. Get the cvpr2016 stuff from https://github.com/reedscot/cvpr2016 and icml2016 stuff from reedscot/icml2016.

To answer your question:

Side note : it seems the default parameters for *_split_path variables links to flowers_icml/*.txt.

Lastly, if you use the provided .hd5 file in the Datasets section, you don't need any of the other files. You only need all the data if you want to convert the data into hd5 file yourself. If you're not able to use the provided hd5 file for version incompatibility issues, try to add this requirements.txt file and install the appropriate versions with pip install -r requirements.

shahhaard47 commented 4 years ago

Lastly, if you use the provided .hd5 file in the Datasets section, you don't need any of the other files. You only need all the data if you want to convert the data into hd5 file yourself. If you're not able to use the provided hd5 file for version incompatibility issues, try to add this requirements.txt file and install the appropriate versions with pip install -r requirements.

Lmk if this helps!