Hello all. I’ve only made it through lesson 3 so far, so feel free to ignore me if this is covered in future lectures. I have approx 10,000 images of houses and their listing prices. Just for fun, my goal is to create an image regression model starting from resnet34 (similar to lectures but not using binary/multi-label classification). I’ve succeeded in building the learner but have run into a problem…

Unfortunately, I have extremely high training and validation loss (massive under-fitting?). I’m using `lr_find()`

to capture a learning rate slice like in lectures, but the method provides a pretty big learning rate for my problem. I first tried 1-10 and then 1-100. Both provide pretty large values for training and validation loss. I’ve used `.normalize(imagenet_stats)`

to normalize my data to the resnet34 model and verified the learner is using MSE for the loss function. I’m not sure what I’m missing but I have a feeling it’s something very fundamental and simple. See snips below…

Any help is greatly appreciated!