Tobias Hinz

Hi, there are some images for which SinGAN may perform better than ConSinGAN. ConSinGAN tends to perform better for images with more structure in them. In general, the easiest thing...

Hi, I suggest increasing the number of `--trained_stages` (default is 6) in general. This should already improve results for higher resolutions. If this is not enough you can also increase...

`--min_size` should be left at some small value, so I wouldn't change it or at least not increase it. `--min_size` and `--max_size` do not need to have a proportional relationship,...

Without seeing the code it's difficult. Have you changed how the parameter `--gpu` is handled (in the main_train.py file)? By default it's set to 0 and later in the code...

Sorry for the late response. What kind of GPU are you running this on and how much VRAM does it have? I run all of my experiments on a single...

Hi, we simply change the learning rate for different layers in the generator. This is done [here](https://github.com/tohinz/ConSinGAN/blob/8b4681fc47796dacf0b229e50214b2c36fec860b/ConSinGAN/training_generation.py#L97).

Can you give me a print out of the `state_dict`? For a quick fix try `netG.load_state_dict(state_dict["netG"])` to see if that works.

Thanks for spotting this, I have just fixed this in the code.

Hi, the method you're trying to call is from the original AttnGAN implementation and is used to generate images from "novel" sentences (i.e. not from the validation set). To avoid...

The dimensions should be the following (num objects = how many objects you want to specify for the object pathway, in our case usually 3): bbox: [batch size, num objects,...