Slow tumor generation

[Luffy03]

Thanks for your excellent work first! I try to use your online transform in main.py directly. However, the speed is too slow and the training is stucked in the initial step. I want to know whether this situation happens in your experiments?

[qixinhu11]

Thanks for your interests in our work! It indeed takes some time and CPU memory to initialize the synthetic textures (I save them to the memory) and cache the data, usually less than 10 mins. But the training time will be fine during other epoch. There might be some smarter ways to achive this, you are more than welcome to drop some insights.