Danila Rukhovich comments

Results 114 comments of


                                            Danila Rukhovich

RuntimeError torch.cat(): expected a non-empty list of Tensors

Hi @advancing-panda , Did you make any modification in the code? Looks like this error can happen with the very low probability if all 16 scenes per batch have no...

Testing Using Pretrained Model Results in Different Performance Across Multiple Runs

Hi @YunzeMan , We didn't meet nondeterministic issues on ScanNet. Can you please share `.log` file of this run? Also may be check something from [torch randomness guide](https://pytorch.org/docs/stable/notes/randomness.html)?

Testing Using Pretrained Model Results in Different Performance Across Multiple Runs

No, this code for sure is able to [reproduce](https://github.com/SamsungLabs/tr3d/issues/3#issuecomment-1531389211) our metrics... May by you can try our implementation in mmdetection3d [codebase](https://github.com/open-mmlab/mmdetection3d/tree/main/projects/TR3D)?

Testing Using Pretrained Model Results in Different Performance Across Multiple Runs

Btw, i think i understand this little randomness in test stage. Here in SparseTensor [construction](https://github.com/SamsungLabs/tr3d/blob/main/mmdet3d/models/detectors/mink_single_stage.py#L67) the default `quantization_mode` is `RANDOM_SUBSAMPLE` following [MinkowskiEngine](https://github.com/NVIDIA/MinkowskiEngine/blob/master/MinkowskiEngine/MinkowskiSparseTensor.py#L130C88-L130C104). Can you try with `UNWEIGHTED_AVERAGE` here?

Unusual Z-Value Predictions and Multiple Predictions per Class on Custom Dataset

> Achieved mAP@50 of 0.76 and mAP@25 of 0.87 Looks like the metrics are very good. So these large boxes should have low scores. What scores do they have?

Unusual Z-Value Predictions and Multiple Predictions per Class on Custom Dataset

I see you have great metrics on all classes except these `Base right ear` and `Base left ear`. Do you have them in your training data? Are their gt boxes...

Unusual Z-Value Predictions and Multiple Predictions per Class on Custom Dataset

Try to play with `iou_thr` in your config. This is NMS parameter regarding how much 2 boxes should intersect to be recognized as duplicates.

CUDA Out of memory during training on S3DIS

Hi @Yrviocnhnie , For sure we did all our experiments on S3DIS on a single 3090 gpu. Have you checked that your pytorch, cuda and cudnn versions are not least...

question

Hi @mcmingchang , We didn't consider rotation in our paper as ScanNet and S3DIS datasets consider axis-aligned bounding boxes. However our code can be easily updated for rotated case, as...

One line that simply crashes all RAM memory

thank you for quick response. i'm inside docker with ubuntu 22.04.3. the host os outside docker is ubuntu 20.04.6 ``` Package Version ------------------------- -------------- accelerate 0.31.0 addict 2.4.0 aiohttp 3.9.5...