OpenPCDet

OpenPCDet copied to clipboard

how to evaluate on custom dataset?

1

I have successfully trained models on my custom dataset, however, when it went to the test phase, it showed me the following error, i thought the reason is custom_dataset.py doesn't...

InfiniteSamele

Different result between paper & PCDet code in kitti dataset (Pointpillar)

2

hello, i don't konw why the test result of PCDet 's Pointpillar have a little big improvment above the paper. For example, use the model pth in PCDet have the...

Townjj

ValueError: need at least one array to stack

1

At present, I only marked 5 point cloud data, and the prompt Database filter by min points vehicle: 5 => 0 appears during training, resulting in ValueError: need at least...

JulyLi2019

target assign cost too much，and gpu util is low, how to improve it ?

1

how to improve target assignment operation when set batch_size = N like `15`, class num = 6 or more bigger, function `assign_target_single` will be called 90 times , and it's...

chenrui17

Can we Export the model to standard Pytorch Model format ?

4

I wanted to know if you can export the models to standard pytorch models ?

ammaryasirnaich

When I use " POINT_CLOUD_RANGE: [-40, -134.4, -2, 40, 70.4, 2]" and set batch_size=8 in KITTI, training is ok, but when I change the batch_size=12, it outputs the following errors:...

beyondpzk

I want to count TP, FP, FN

2

Hi. I want to know how to count true positive, false positive and false negative in the eval.py. Thanks.

frogbam

Problem about visualization, why do these point clouds appear in two directions？

2


eucww

Fail to download the pretrained models

1

The first step of `Quick Demo` is : Download the provided pretrained models as shown in the [README.md](https://github.com/open-mmlab/OpenPCDet/blob/master/README.md). Then all the pretrained models are stored in `gdriver`. But, `Hmmm… can't...

ausk

multimodal model / pipeline

11

Hi All, Thank you for this wonderful repo! Do you support multimodal models (point clouds + RGB)? if not do you have a pipeline / dataloader we can use for...

AvivSham