xuanyuzhou98/squeezesegv2

Implementation of SqueezeSegV2, Improved Model Structure and Unsupervised Domain Adaptation for Road-Object Segmentation from a LiDAR Point Cloud

squeezesegv2's Issues

How to use the code for own dataset

I do have converted dataset from VLP16 lidar. but code says its currently it supports only KiTTI dataset.

how can i run the code for my own data set??

Please help me

could you open source for the kitti data process code?

How much GPU memory is needed?

I try to trian the model with one GPU (TITAN XP with 12G memory) and get tensorflow error: Resource exhausted: OOM. Thus, I want to know how much GPU memory is needed for training and evaluating respectively?
Thanks!

No module named 'kitti_squeezeSeg_config'

there is a strange error that are shown in the code when running the demo.py since it says that

SqueezeSegV2\src\config_init_.py", line 1, in
import kitti_squeezeSeg_config
ModuleNotFoundError: No module named 'kitti_squeezeSeg_config'

even though the file are there

Question about the geodesic loss

Hi, Bichen Wu.
Firstly, thank you for your brilliant work! I looked up your code but didn't find the implement of geodesic loss. In which part is the implement of this loss? Please help.

tensorflow.python.framework.errors_impl.DeadlineExceededError: Timed out waiting for notification

Hi,
I have downloaded this code and am able to successfully run the demo.py file. I am getting the above mentioned error while trying to run the train.sh file as "./scripts/train.sh -gpu 0 -image_set train -log_dir ./log/". Is someone else also getting the same error? I am running it on a GEForce RTX 2080 Ti with a single GPU

where is the domain adaptation part?

In your paper, you add a Geodesic Correlation Alignment and Progressive Domain Calibration to deal with domain adaptation. But I can not find it in your code. Could you provide that part of code?

Learning rate changes abnormally

Hi! Thank you for sharing your work. When I look at the training effect in the tensorboard, the learning rate shows the following trend. The learning rate change is different from the exponential decay trend set in the code. Can you give me some suggestions?

Could you open the synthetic raw dataset without being projected to 2D?

Hi, thank you for sharing this outstanding project. It is very kind of you to provide the link for downloading the projected dataset. However, I need the synthetic raw dataset without being projected to 2D. Could you open the raw dataset?

How to train on GPU?

Training on GPU is not working for me by setting -gpu 0,1,2 command line options.

./scripts/train.sh -gpu 0 -image_set train -log_dir ./log/

I'm running SqueezeSegV2 on a conda virtual environment with tensorflow-gpu version 1.4.1

$ pip list | grep tensorflow
tensorflow-estimator 1.14.0
tensorflow-gpu 1.4.1
tensorflow-tensorboard 0.4.0

By invoking the training script the GPU remains mainly unused

$ nvidia-smi
Thu Aug 1 15:50:58 2019
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 430.26 Driver Version: 430.26 CUDA Version: 10.2 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce RTX 2070 Off | 00000000:15:00.0 Off | N/A |
| 29% 30C P8 14W / 175W | 107MiB / 7982MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 1 Quadro P1000 Off | 00000000:21:00.0 On | N/A |
| 34% 40C P8 N/A / N/A | 507MiB / 4030MiB | 4% Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 26958 C python 93MiB |
| 1 1599 G /usr/bin/gnome-shell 92MiB |
| 1 2379 G /usr/bin/gnome-shell 392MiB |
+-----------------------------------------------------------------------------+

I expected GPU training to be working out of the box. What am I missing?

GTAV dataset

Hi, thanks for sharing this very interesting approach !

Is it still possible to get the GTAV dataset ? I heard that there were some issues with TakeTwo...

Confused on 'We name the root directory as $SQSG_ROOT.'

Hey everybody, I am a newbie with Ubuntu but I tried searching Google for this, I'm confused on what I'm supposed to do after cloning the repository here.

How am I supposed to rename the root directory as $SQSG_ROOT? What does that even mean?

I'm researching online too obviously but could somebody link some documentation or something to explain this? Thanks!

How to train on kitti dataset?

How to train on kitti?

To be specific, there 2 questions:

1. What does network input which is inside samples directory, it is a array of [64, 512, 5], but normally point cloud is 4 dimensions, where does the 5th dimension from?

2. How to generate lidar-2d data?

how to freeze the pb for c++

if i use the typical freeze_graph.py method, it cannot produce the placeholders. I wonder if the fifoqueue causes this question. could you share the frozen script

Citing errors

Could you check your citing format again? I get errors when cite by latex in bib format. Thanks!

A question about KITTI raw dataset

Thanks for sharing your work!
I have a problem about your lidar_2d data, which only has 10848 .npy files. But there are 12915 files raw data with tracklets in KITTI dataset.
Do you throw away some files? Could you verify this problem?
Thanks! Waiting for your reply.
The same issue is here: BichenWuUCB/SqueezeSeg#35

Synthetic Dataset Access

Hi @xuanyuzhou98,

I submitted the data request several days ago and I still have not heard back. Is the dataset available somewhere to download or are there previous copies?

Thank you very much.

Best,