liushihhung / gicn Goto Github PK

What is new_bonet? In train.py there is a file being copied with that name but it is not in the repo

How can i use semantic net in your project?

The semantic net was used in center selection mechanism in the paper, but in the code this part is commented out and no weights.

Hi,

I was wondering if the scripts to generate the center heatmaps are also available? Thanks.

can not find them in this repo

Hi,

I've created a Docker container with all the dependencies listed in your repo. However, when I run the training process I get the following runtime error:

Traceback (most recent call last): File "train.py", line 409, in <module> bat_pc[:, :, :3], bat_pc[:, :, 3:9], global_features, sidx, 0 File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py", line 550, in __call__ result = self.forward(*input, **kwargs) File "/usr/local/lib/python3.6/dist-packages/torch/nn/parallel/data_parallel.py", line 153, in forward return self.module(*inputs[0], **kwargs[0]) File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py", line 550, in __call__ result = self.forward(*input, **kwargs) File "/root/gicn_model.py", line 316, in forward grouped_points = query_and_group(xyz.contiguous(), new_xyz.permute(0,2,1).contiguous() , points.permute(0,2,1).contiguous()) File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py", line 550, in __call__ result = self.forward(*input, **kwargs) File "/root/Pointnet2.PyTorch/pointnet2/pointnet2_utils.py", line 259, in forward new_features = torch.cat([grouped_xyz, grouped_features], dim=1) # (B, C + 3, npoint, nsample) RuntimeError: CUDA error: device-side assert triggered

Hi, do you have the record of how your final loss looks on the S3DIS dataset? For example, for the pre-trained model you gave after 75 epochs, what's the magnitude of those different losses?

Thanks a lot for sharing the amazing code. I'm wondering if it can be run on Windows or which part of would need to be modified to run on Windows.

Thank you for your help in advance!

Hi,

I've been trying to test your code but I can;t retrieve results on level which is declared in readme and article.

Each time I do the evaluation result are on level of 1-2% in prec and less than 1% in rec. I've tried to run main_eval on trained model included in repository, on s3dis dataset available in jsis3d repository. (zip which can be downloaded according to your readme file)

Like others I've met several issues during code setup, what made me do some minor changes, but I have no idea if these had influence on results which I receive.

So here is the short list of changes in code and reason why I did it:

1. "Unable to open object (object 'semIns_labels' doesn't exist)" - I have change the names of column to labels, since in jsis3d database has no such column: semIns_labels

2. ModuleNotFoundError: No module named 'new_bonet' - since new_bonet doesn;t exist I have changed the import to gicn_model

3. "Unable to open object (object 'pc_indices' doesn't exist)" - I have change reading of this column into arange in following manner:
   pc_indices = np.arange(block_id4096, block_id4096 + 4096)#fin['pc_indices'][block_id] # [4096] due to fact that, this column doesn;t exist

4. input_bat_pc_indices = copy.deepcopy(bat_pc_indices[0].squeeze(1))
   numpy.AxisError: axis 1 is out of bounds for array of dimension 1 - what I did was deleteing the squeeze, since array has a single dimension

5. FileNotFoundError: [Errno 2] No such file or directory: '/media/marcin/StorageForData/GICN-master/./data_s3dis/SCN_prediction/area5_2/WC_1.txt' - I have change the path to area5 without _2, since such directory doesn;t exist.

6. pc_indices_raw = scene_result[block][0]['pc_indices_raw'][0].squeeze(1)
   

ValueError: cannot select an axis to squeeze out which has size not equal to one - I have changed squeese to axis 0 (squeeze(0)), since this is the only axis possible to squeeze and without squeezing it I fell on another issue.

7. ModuleNotFoundError: No module named 'helper_data_plot' - I have simply change the import to data_plot since helper doesn;t exist

Could you give me some hint or help on the reason of such unexpected prec and rec results? Any help will be much apreciated.

Best Regards

Thanks for your work. I notice that the key 'pc_indices' appears both in train and inference script, but missing in the h5 files.

Hello,i noticed that there is a (6,3) shape array named "instance_sizes" in train.py，what does it mean? please

Thank you for your great working. Could you please share with us pretrained model ?