Oliver Groth

Hi @phongnhhn92 , I haven't really looked into multi-GPU training, yet. But I think, you'd be able to specify the parameters for it in the [GPU options](https://github.com/ogroth/tf-gqn/blob/master/train_gqn.py#L131) of the estimator.

Hi @PengchengWang , your observation is correct. I've experienced similar behavior when I trained on rooms ring. The model learns to render the quadratic rooms (i.e. floor, background and sky)...

@PengchengWang @wenbingz I've uploaded model snapshots for the Shepard Metzler datasets, together with tensorboard summaries, run commands and hyper-parameters. Please feel free to check them out [here](https://github.com/ogroth/tf-gqn#model-snapshots).

@loganbruns That looks great, thank you for sharing these results! :) I'd be very happy to include a data loader for the CLEVR dataset (either from raw files or from...

Hi @loganbruns , the new input pipeline is now in master. Would you mind modelling your input_fn for CLEVR after [this one](https://github.com/ogroth/tf-gqn/blob/master/data_provider/gqn_provider.py#L189)? Also, you can include data generation and conversion...

Hi @curryJ, sorry for the long silence, I've been busy with other projects. I'm currently in the middle of re-writing the input data pipeline (see commits on the dev-branch) and...

Hi @curryJ , the new [input_fn](https://github.com/ogroth/tf-gqn/blob/master/data_provider/gqn_provider.py#L189) should be able to handle all GQN datasets. Also, both training script and input_fn now come with img_size parameters to resize images from different...

As I've explained [here](https://github.com/ogroth/tf-gqn/issues/37#issuecomment-526115035) the loss seems to plateau because of the reconstruction term having reached the perfect means and the only thing which is optimized afterwards is the KL...

The self-attention encoder you've added in gqn_encoder.py looks neat. Can you send me a pull request for that on the [attention_encoder branch](https://github.com/ogroth/tf-gqn/tree/attention_encoder)?

Hi @waiyc, how do you feed your custom dataset? Have you written a new input_fn or at least made sure that the data structure you're feeding complies with the [TaskData](https://github.com/ogroth/tf-gqn/blob/master/data_provider/gqn_provider.py#L24)...