superhans / colorfromlanguage Goto Github PK

Great work! I'd like to replicate the results and try out new ideas. Unfortunately, I couldn't have managed to download the data shared on the readme file,

https://obj.umiacs.umd.edu/learning_to_color/coco_colors.h5

If it's possible, could you please fix the link / share the data?

@superhans Thanks for providing the code.

What is the test data that you used in the paper to report results? Could you provide any evaluation scripts as well?

Thanks.

Nice work! Could you share the requirements file as well? Which version of python and torch do you use?

Hello, you mentioned in your paper that the model is pre-trained on ImageNet.Do I need additional training on ImageNet before following your instructions?Alternatively, Resnet101 generates image_features. H5, which represents the process of training the model on ImageNet.

Hello Sir,
I am a beginner and I liked your work. I was reading your paper. Can you please give a brief about the codes and also the models to me. Do the code contains both FiLM and CONCAT based models or is it for FiLM only. Also, sir is the model provided pretrained over ImageNet. Please help me out. If you have your original models as described in the Paper. Can you please share it with me? It will be very helpful.

Interesting work!

When I was trying to run the model, I found the w2v_embeddings_colors.p file missing from the priors folder. I wonder if this is a word2vec pre-trained embedding, or did you train the embedding yourself?