Pixel2Mesh
Pixel2Mesh copied to clipboard
Quality of the images for input
Hi, I was trying to get 3D shapes from images download from internet. But the results were not good. Can you specify the quality of the PNGs that needs to be fed ? The background colour ? Or any specific details to get good results ?
Can this work on 3D people dataset or Deepfashion 3D dataset ?
Hi, Do you know how to download the dataset and unzip it?
Me too. When I took a screenshot of the png of the car in examples, when I fed that car into the model as a different image that I took a screenshot of, it thought the car was a couch. Previously it predicted a car quite accurately but seems to like to predict furniture. There must be something I am missing on png settings for this to happen.
I am also getting vague 3D model results. So, are there any updates on this?