skanti / scan2cad Goto Github PK

Is it easy to release the baseline methods that you are comparing against here http://kaldir.vc.in.tum.de/scan2cad_benchmark/?

Thank you!

Best,
Kaichun

Hi, skanti，I have filled out the form, could you please send me the dataset for research purposes.

as the title, I run the ./run.sh follow you step after i download the scan2cad_download_link.zip, but when the training process begin, the cuda memory if slowly increase from 1500M utils process terminated , and i will get a out of memory error

Hi,
I received the email but the .zip download link doesn't work. Thanks a lot for any help!

Hi , I’ve filled out form but have not received a response. Is the dataset still available ?

The dataset faces mainly consist of 3 vertices hence the array f is always a 3-tuple array, however in line 115 of Annotation2Mesh.py it just call f[3], which makes the program crash.
Did you notice this apparent error or is there something I missed to cause such problem?

I have downloaded the scannet and shapenet data. I run the Annotation2Mesh.py line 145 of Annotation2Mesh.py cad_mesh = pywavefront.Wavefront(cad_file, collect_faces=True, parse=True), which makes the program crash. It raised ValueError ValueError: Trying to merge vertex data with different format: T2F_N3F_V3F.Material material_1_24 has vertexformat. Could you please help me to solve this problem. Thank you a lot.

hey guys, sorry for posting here but can you send me the scan2cad dataset for research purposes i have already filled the google forms .

Sorry to open an issue here, I've already filled the google form that requests for this dataset (for research purpose), could you kindly help check the status? Thank you in advance!

code compiles fine for CropCentered.so but still this error..when i try to run generate_correspondence.py..any cue would be of great help.

The real CAD modes are provided as part of the input or retrieved from a model dataset like shapenet?

Hi,

Is it possible to share the actual data generation script used in the paper with us (which includes both positive and negative sample generation)? Thanks!

Can we also get access to the annotation tool?

Hi,

Thank you for sharing this impressive project. I have one question: how did you visualize the generated .vox files? What software/library did you use? Thanks in advance.

hi, I want to "generate the df voxel grids for the CADs we used this repo", but I can not get the right data, comparing with you provided. And how can I get the grid2world matrix?

First of all, thank you for sharing the Scan2CAD code online. I tried to reach you via email, however I haven't received any replies yet, so I'm going to post my questions here.

1. Do you by any chance have a pre-trained network available for testing?

2. In your paper, you mentioned that you generated both positive & negative training sampled. In the github repository, it seems like you only have code for generating positive samples. Can you share the code for generating negative samples as well? In addition, in the paper you also mentioned that you used a pre-trained CAD encoder. Is the pre-trained encoder included in the repo?

3. Related to the previous question, is the code in git repo enough for reproducing the results in the Scan2CAD paper, or is there more that needs to be added? Is there an evaluation script that I can use to reproduce the same results as shown in the table in your paper?

4. What generated data files contain those correspondence information generated by your network?

Thank you in advance!

Hi,

Thank you for the excellent work! I read the paper recently and found that the baseline methods use the patch around the harris keypoints as input. So did the method proposed use the patch around the harris keypoints as input either? If yes, so can i ask what the parameter you use to get the harris keypoints? i try the Harris3D by the software PCLKeypoints but the detector perform bad. I will appreciate if you can give me some hint. Thank you!

How can I get the .vox data for scannet scene? Do you have the code or the data?

Hello, nice work! But I'd like to know where to find the evaluation script that do evaluate on the retrieval results. Follow http://kaldir.vc.in.tum.de/scan2cad_benchmark/documentation, I can't find the script in this repo. I can only find the evaluation on the 'heatmap' and 'ismatch'. Maybe I just missed that?

When I visit the link: https://kaldir.vc.in.tum.de/scan2cad_benchmark/, I got the following message:

Not Found
The requested URL was not found on this server.

Apache/2.4.52 (Ubuntu) Server at kaldir.vc.in.tum.de Port 443

hi, thank you for your constructive work. I'd like to run the code and have applied for the scan2cad dataset, but never recived feedback, and the google-form seems not working anymore. could you please provide me some other channels？

Hi! I've filled the form for requesting dataset at least one week ago, but I received nothing in my mailbox. Maybe there's something wrong with the google form. Could you please send me the dataset? I really need this for research purpose. Thank you very much!

Hey, thank you so much for creating the dataset! I have filled the google form to request the dataset but did not receive any response so far. Could you kindly help check the status? Thank you in advance!

hi
I use volumetric fusion get the TSDF result of ScanNet(0470_00), the voxel resolution is 3cm and truncation is 15cm(grid num(116,95,46)). The result is very worse, can you give some more detail about obtain TSDF of ScanNet datasets.

Hello, I'm working on training the network to predict cad alignment on RGB-D scan. In my understaning, the annotation for testset is not provided to prevent the parameter tuning on testset. But is it the same with validation set, either? In main.py script, it seems that a scene list for validation is required but I cannot find it. How can I handle overfitting on train set?