simao

@animebing, could you make multi-gpu to work?

I can load the model from the Checkpoint (as below), and get predictions from it. However, I **can only get predictions for an output_tensor before the ROIAlign layer**, but not...

@XieSufe, images is just any list of numpy arrays you want to feed to the graph. As an example just do ```python import PIL pim = PIL.Image.open('/dir/to/image.jpg') nim = np.array(pim)...

I can load the model from this Checkpoint (https://github.com/CharlesShang/FastMaskRCNN/issues/86#issuecomment-309022377), and get predictions from it. However, I **can only get predictions for an output_tensor before the ROIAlign layer**, but not for...

@xwjabc Can you explain the value lambdas you used for each of the different losses?

@qin79 , The `make_bu_data.py` script parses the [mscoco train/val/split object detections](https://github.com/yahoo/object_relation_transformer/blob/master/scripts/make_bu_data.py#L27-L30) generated by the [bottom-up attention codebase](https://github.com/peteanderson80/bottom-up-attention). Those files, only include the following fields `['image_id', 'image_w','image_h','num_boxes', 'boxes', 'features']` so there...

@HN123-123 were you able to fix this bug?

Thanks Nikolai, On a single V100 GPU the extra 30 epochs of self-critical training should take 3.5 days. I believe you shouldn't run into any memory error. Please let us...

@jiajunhua You should be able to run the current version of the codebase in Python3 (e.g. Python 3.6.8). Please try it out and let us know if you run into...

@luo3300612 Thanks for your observations. Equations (6) and (7) in the paper, show that indeed the box_embedding `Emb(\lambda)` is just a function of the bounding box displacements, and therefore constant...