Scan2Cap issues

Some puzzles about dataset processing

5

I once encountered a problem when preprocessing the scannetv2 dataset. I tried to solve this problem, but I'm not sure whether my solution is reasonable. I'd like to discuss it...

yvfengZhong

different experiment settings between training end to end scan2cap and the fixed-detector scan2cap

In end-to-end scan2cap, the relation graph's input is the origin proposals without nms. However in fixed-detector scan2cap, the relation graph's input is the origin proposals with nms. I think that's...

ch3cook-fdu

About output info "error occurs when dealing with graph, skipping..." during training

Thanks for your great work in the 3D dense caption! I found the info "error occurs when dealing with graph, skipping..." occasionally emerges during the training phase. I am wondering...

SxJyJay

Update load_scannet_data.py

The aligned_instance_bboxes is computed using axis aligned mesh vertices, which means it is different than the bboxes computed using vertices from the original mesh(.ply) file. So aligned_instance_bboxes should be passed...

databaaz

How could we get folder render-based/ used in Scan2Cap-2d

2

Hi Scan2Cap team, I've mentioned that several folders like "projected-based/renders" "render-based/renders" appeared in the conf.py, but I didn't find these images from ScanNet, ScanRefer or Scan2Cap. Did we use some...

YuiTH

How to learn relative orientations

Hi,your work is very great. Could you explain indetail why you use "CAD model alignment annotations" to learn the object relative orientations in the relational graph ? I would appreciate...

cuiheng1234

How did you measure the accuracy in Table 3?

Hi Dave, may I ask how did you measure the accuracy in category, attribute, and relation of the generated captions as presented in Table 3 from the main paper? Thanks...

xwhjy

Request for checkpoint of end-to-end training experiment

I'm now training the Scan2Cap model in an end-to-end manner. Could you please release the training log and checkpoint for the end-to-end training dense caption model?

ch3cook-fdu