val-iisc / 3d-lmnet Goto Github PK

Hi

A few papers about point clouds reconstruction gave their result on multi categories. DeformNet (here) gave their results and their experiment on PSGN. They got result of PSGN like 0.13 (CD) while you got 0.05 (Table 3 in your paper).

I've thought this before. PSGN originally used multi categories to train their network and DeformNet use PSGN's network directly to test certain single category. I assume that you use single category to train PSGN network and use it to test that category. So DeformNet's result of PSGN is much higher than yours.

But Dense 3D Object Reconstruction (here) shows a result of PSGN like 0.028 (CD of airplane) instead of your 0.037 (CD of airplane). I also implement PSGN several times with several methods and got similar result (0.028).

I suppose that your paper didn't mention that how did you get your result of PSGN or your normalization about your point cloud and so on.

My confusions are here and looking forward to your reply.

Hi, I'd like figure out some details about how you evaluate PSGN.
The prediction (scaled, N x 3) from PSGN is tranformed by the following rotation matrix,

then icp is applied between pred and gt for finear alignment. The final score is computed between the transformed pred and gt.
Is my description of the evaluation process correct?

Sorry to bother you again. You said that you will provide the ground truth point clouds before but now due to memory restrains you give ShapeNet meshes instead. I totally understand that but could you release your point clouds generation method from meshes and how to package your rendered images, point clouds together to get training started? I supposed that here isn't clear enough to use your dataset to start training.

An inspiring work.
I want to know the specific process of extracting the gt point clouds from the provided meshes on Pix3D dataset, in particular, the coordinates transformation from original meshes to the processed point clouds. Whether you just normalized the coordinates of these sampled points, or utilizing the camera's extrinsics and intrinsics to transform these sampled points to image plane?
If possible, could you please offer the corresponding specific preprocess code?
Thanks a lot!

Hello! your idea is novel , and the result has great improved , but I cant recieve the reconstyructed point clouds of PSGN, could you provide how to visualize the PSGN result like your paper shows ? Thank you!

Hello, thank you for your excellent work. I have a question about the dataset preparation. Say if we have access to the depth image and camera parameters (intrinsic and extrinsic), we backproject depth image to 3d space using intrinsic matrix and transform it from camera frame to world frame using extrinsic matrix. Do sampled ground truth point cloud and point cloud recovered from depth lie in the same frame? Or the ground truth pc lies in their own body frame?

hello, when I run the bash scripts/train_plm.sh, there exit a question:
Traceback (most recent call last):
File "train_plm.py", line 168, in
z_mean, z_log_sigma_sq = image_encoder(img_inp, FLAGS)
File "/home/user05/.conda/envs/tf/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 442, in iter
"Tensor objects are only iterable when eager execution is "
TypeError: Tensor objects are only iterable when eager execution is enabled. To iterate over this tensor use tf.map_fn.
could you tell me how to solve, thanks .

Sorry to bother you . I downloaded your trained model,and use
bash scripts/metrics_shapenet_lm.sh
to evaluate the Chamfer and EMD，but the values are too low(eg:airplane:Chamfer: 0.001360 and EMD: 0.001943).I want to ask what might be willing to cause this？The Pix3D‘s result is normal.
Looking forward to your reply!

Hi, I want use the evaluation to see the cd, emd etc. But it report the issue. I see some file but didn't find the detail. I miss something here?