Preparing Your Dataset For Training When training a multi-class segmentation model, loss functions such as the CrossEntropyLoss() in Pytorch require the mask to be categorically labeled. As in the image above, the RGB mask (middle) is properly segmented but all…
Clustering Depending on your data and objective, you may not even have to train a deep-learning model for image segmentation. Here is one way to apply segmentation using Kmeans clustering. Implementation from sklearn.cluster import KMeans from matplotlib.image import imread import…
Objective DAFormer is an architecture proposed to improve domain adaptation for segmentation models. For the encoder, the “Hierarchical Transformer” is used due to being revealed to be robust to domain shifts. The decoder applies context-aware fusion which utilizes domain-robust context…
Unsupervised Intra-Domain Adaptation CNN-based neural networks heavily rely on annotated data which are labor intensive. To cope with this limitation, automatically generated data are used to train segmentation models. However, models trained with these data struggle to generalize to real…
Image Segmentation Here is one way to implement Image segmentation using contour detection. The precision is not quite there, but I was amazed that you can do segmentation just with contour detection. I think it is still usable if.. You…
Simple SGD vs Cyclic Learning Rate I compared the training speed between two optimizers by training a UNet Model. Simple SGD optimizer = torch.optim.SGD(model.parameters(), lr=0.01) Cyclic Learning Rate optimizer = torch.optim.SGD(model.parameters(), lr=0.01) scheduler = torch.optim.lr_scheduler.CyclicLR(optimizer, base_lr=0.1, max_lr=1e-4) Cyclic Learning Rate…
Polynomial Learning Rate For deep learning models, the learning rate is one of the most important hyper-parameters in any deep neural network optimization process. Polynomial Learning Rate is a proposed technique to apply learning rate decay and optimize such process.…
Today I tried to do transfer learning using PIDNet (Since I just learned about PIDNet). Compared to my first attempt, the output is getting slightly better but still not to the level where it is actually useful.
Today I’ve learned about bilinear interpolation for images, so I’d like to share it here. To simplify the concept, here is an example if we were to upscale an image by a factor of 2 using bilinear interpolation. For semantic…
torch.nn.CrossEntropyLoss() Using torch.nn.CrossEntropyLoss() as a loss function for semantic segmentation models was first confusing for me, so I’d like to share it here. CrossEntropyLoss is for multi-class models and it expects at least 2 arguments. One for the model prediction…