tansor / vecroad Goto Github PK

Recently I read your paper VecRoad. It is a good job and quite inspiring. But I have some questions about the proposed method and hope you can give me some hints:

1. In 3.2. Iterative Exploration, you added the Trajectory Exploration in VecRoad, and it seems that this is critical for satisfactory performance based on ablation studies. However, the explanation is too concise, and I do not understand the connection between your training strategy and experience replay. From my perspective, you generate multiple predictions instead of one in each step. Why does this work? How could this strategy relieve the drift problem in trajectory prediction tasks? And why is this related to experience replay, which is a commonly used RL strategy?

2. May I know the overall training time of VecRoad in your task? Considering that the iterative decision-making process is always with high time consumption.

3. Do you have any plans to release your training code in the future? Or it cannot be published at all?

Thanks for your patience and really appreciate it if you could offer me any help.

Hey there!

I belong to an open source security research community, and a member (@oivrip) has found an issue, but doesn’t know the best way to disclose it.

If not a hassle, might you kindly add a SECURITY.md file with an email, or another contact method? GitHub recommends this best practice to ensure security issues are responsibly disclosed, and it would serve as a simple instruction for security researchers in the future.

Thank you for your consideration, and I look forward to hearing from you!

(cc @huntr-helper)

Hi, the link provided to download pretrained model files and the dataset is not working for me, perhaps it does not work in my region. Can you please upload that files on google drive also?
I'll look forward to your response. Thanks.

Hi, thank you for your code sharing. When I ran your inference code, I found that in your DecoderBlock, you use an attribute called cat_ch, which is not defined. It may be a small mistake.
The output is as below.
initializing model Traceback (most recent call last): File "infer.py", line 554, in <module> main() File "infer.py", line 54, in main net = prepare_net().eval() File "infer.py", line 430, in prepare_net net = RPNet(cfg.TRAIN.NUM_TARGETS) File "/home/molijuly/github/VecRoad/model/model.py", line 154, in __init__ DecoderBlock(self.ft_chs[0], 32), File "/home/molijuly/github/VecRoad/model/model.py", line 39, in __init__ ConvReLU(self.cat_ch, self.res_ch, 3), File "/home/molijuly/anaconda3/envs/pt10py37/lib/python3.7/site-packages/torch/nn/modules/module.py", line 576, in __getattr__ type(self).__name__, name)) AttributeError: 'DecoderBlock' object has no attribute 'cat_ch'

I want to extract centerline in some raster data. Could you tell me how to do?

I download the Pretrained models from the Baidu NetDisk,but it seems wrong to open it or decompress it. Could you tell me how to fix this problem.Thanks very much!

Hi, thank you for sharing the code. I'm using it to do my research. I have a few questions about it.

1. What's the meaning of the param is_key_point in the "push" method in model_utils.py and how to set it?
2. How does "proc" method work in the "push" method?
   I'm appreciated if you could add more annotations to this piece of code.

First of all, I am very grateful that you have open sourced the code. Here I have a small question to ask you whether there is an array overflow when the file--infer.py is interrupted here.
My GPU memory is 24G, and the virtual environment is configured according to your instructions. Waiting for your answer, thank you very much.

Thank you for your excellent work! But I encounter a problem when I run infer.py. The problem is when I set INFER_STEP: "start", START_FROM_ROAD_PEAK: True, the file can successfully run. If I change the INFER_STEP to"after_seg", it will report an error when doing post_process_graph. Can you give me some advice? Thanks again.

infer anchor start
batch_size:1
Except save graph yunnan_city
Traceback (most recent call last):
File "infer.py", line 586, in
main()
File "infer.py", line 222, in main
post_process_graph(graph_dict)
File "infer.py", line 425, in post_process_graph
road_segments, _ = graph_helper.get_graph_road_segments(g)
File "/f2020/kangkai/VecRoad/lib/graph.py", line 635, in get_graph_road_segments
for edge in graph.edges.values():
AttributeError: 'NoneType' object has no attribute 'edges'

 Thanks for you nice work :). 
 I run the infer.py on test dataset and the results are very good. But when I run train.py, the error is:

self.tiles = Tiles(self.cfg.TRAIN.PARALLEL_TILES, pytiles_path=cfg.DIR.PYTILES_PATH,
AttributeError: 'EasyDict' object has no attribute 'PYTILES_PATH'

PYTILES_PATH is not in cfg file and the params seem mismatch with
https://github.com/tansor/VecRoad/blob/master/utils/tileloader.py#L159

Hello,I just see the mask for testing dataset, so can you tell me where is the mask for training dataset, Thank you.

After i skim the papers, i desire to know how to implement the detail of th method.

This is a nice work, well done! But it lacks enough details for re-implementation, such as your training pipeline, how to generate labels, etc. At this stage, it is impossible to implement it without neither the training code nor enough implementation details. Could you please provide more information on training VecRoad?