image classification - deep learning
Budget: $30 – $250 CAD
NO TECHNICAL WRITERS PLEASE!
Hi, I have a code (I'd say medium to high level of difficulty) to do deep learning for image classification task in Pytorch. The code is given me a good results when using cross entropy loss (high train/test accuracies). However, when I try to visualize the features in the feature space using tsne, the figure looks very bad. In my task, I have x classes, so I do expect to have x prototypes in the feature space that are well separated.
I cannot understand why, but I'm looking for an experienced deep learning and python developer who could test my code with different losses, (or whatever he/she proposes), such that my features are separable!
So If you have used any of the distance-based losses, that would be great. Note that I already tried to use the contrastive loss but it didn't work. Not sure if I did it wrongly, or something else.
Hi, I have a code (I'd say medium to high level of difficulty) to do deep learning for image classification task in Pytorch. The code is given me a good results when using cross entropy loss (high train/test accuracies). However, when I try to visualize the features in the feature space using tsne, the figure looks very bad. In my task, I have x classes, so I do expect to have x prototypes in the feature space that are well separated.
I cannot understand why, but I'm looking for an experienced deep learning and python developer who could test my code with different losses, (or whatever he/she proposes), such that my features are separable!
So If you have used any of the distance-based losses, that would be great. Note that I already tried to use the contrastive loss but it didn't work. Not sure if I did it wrongly, or something else.