Dave Z. Chen

Thanks a lot for your interest in this repo. Briefly, FC model is an encoder-decoder pipeline without any attention mechanism, while the Att2all is a model equipped with attention module...

Hi @Ze-Yang, 1. This line is for hashing the voxel coordinates after voxelizing the point cloud; 2. The points sampled in the 1.5m by 1.5m by 3m cubes are voxelized...

Hi @hongchenphd, can you try this: 1) right click and copy address; 2) type `wget ` in your terminal?

@spandanagella normally the HITs should appear within one minute.

Hi there, I'm facing the same problem of reproducing the performance in the PointNet++ paper. The best accuracy I had is 59%. I use the code in [https://github.com/charlesq34/pointnet2/blob/master/scannet/preprocessing/collect_scannet_scenes.py](https://github.com/charlesq34/pointnet2/blob/master/scannet/preprocessing/collect_scannet_scenes.py) to preprocess...

Hi @CurryYuan, could you quickly check if you can get reasonable results using the [checkpoint](https://www.dropbox.com/s/nsrbcfeihmh2bhw/D3Net.7z?dl=0)?

Hi @zhuziyu-edward, there seems to be an internal issue with the server. I'm working on it now. Could you probably email me your evaluation results (.zip file) just in case?

I looked up other implementations and they didn't include the validation loss, either. But indeed those evaluation metrics are pretty critical for us to tell the performance of the model....

Tried two ways. One way is generating texts with teacher- forcing while the other is without teacher-forcing but is constrained to have the same length as the target captions. However...

Same issue here. It seems the model always attend to the same area even if I overfit the model on a very small dataset, e.g. 100 image-caption pairs. I actually...