Perception Traffic Light module run failed #15591
Comments
When you launch a node, it keeps running and waiting for input messages. So it is normal in my opinion.
Seems that lane issues are related to HDMap The errors are because it cannot find the pre-trained model as shown by the error messages You should locate this file in your version, simply search the file name. In my opinion, you do not need to launch the entire project for traffic light. You only need several modules, like HDMap and Traffic Light, and Localization. You do not need dreamview or whatever. If you have your own video, you have to have the corresponding map and localization to provide the ROIs, Check this https://github.com/ApolloAuto/apollo/blob/master/docs/06_Perception/traffic_light.md I have replicated the Apollo traffic light detection and recoginition pipeline in python. It can process pictures with specified initial estimations. Check about this https://github.com/boyang9602/TLR-replicate |
|
@boyang9602 I was looking at the Git project you referenced, but I have some doubts about the project itself. I understand that it works by taking an image and some bounding boxes you pass as parameters, and it provides the pipeline's output that would occur in the Apollo project. My question here is: how do you determine which bounding boxes to pass as parameters? Also, there's something I wanted to test based on the video, which is step 4 of the pipeline, the "reviser." This step could determine the current traffic light information based on the prior information it had. Is that possible here? Lastly, you mentioned that certain map and location configurations were necessary. Are those required here as well? |
In Apollo, it is determined using a camera model to project the 3D traffic lights into 2D pictures. It requires the 3D model of the environment (HDMap), Localization, and camera information. You can check the pre-process step described in https://github.com/ApolloAuto/apollo/blob/master/docs/06_Perception/traffic_light.md#pre-process. However, you can simulate this step by adding some transformation of the ground truth boxes (e.g., shift, resize, etc), i.e., making up some projection bounding boxes using ground truth information, as if they are calculated using the HDMap and localization.
My replication does not include the reviser. However, it is simple as described by the doc
Please refer to the reply for the first question. |
System information
apollo.sh configif onmasterbranch:Problem and steps
First, I will explain the purpose of what I am doing. In the second step, I will outline the steps I followed to install the Apollo project, which might have led to an incorrect installation, and consequently, the error could be related to this. Finally, in the third step, I will show the error, the steps I followed, and the issue itself.
What I am trying to do is run only the traffic camera perception process using a custom video. I want to do this to analyze how the traffic camera flow works exclusively and perform tests focused on that. Therefore, my idea is to use a recorded video, simulating a vehicle-mounted camera, and have that video serve as the input for the traffic camera perception process to obtain the corresponding bounding boxes. To achieve this, my plan is to run the perception module, particularly the traffic camera component, and inject the video into that process somehow.
Initially, I followed the prerequisites, which involved installing the GPU drivers, CUDA, LibTorch, Docker, Docker Engine, and NVIDIA Container Toolkit. After that, I cloned the project and started using it from the master branch. I followed the steps that required running the following commands:
Finally, to run it:
Then, I installed the demo and ran it with:
Up to that point, I was able to open Dreamview and see the demo. Since my purpose is related to the perception modules, I installed additional configurations that were prompted during the build, as they needed to be installed separately. These are all the steps I followed prior to running the perception module.
As I mentioned earlier, I want to run the perception module for traffic cameras. Therefore, I followed the guide:
https://github.com/ApolloAuto/apollo/blob/master/docs/06_Perception/how_to_run_perception_module_on_your_local_computer.md
Following the steps outlined there, the only difference in my process was that instead of running ./apollo.sh build_opt_gpu, I ran ./apollo.sh build. Then, in Dreamview, I selected the following options:
setup-mode: Mkz Lgsvl
car: Mkz Lgsvl 321
map: Sunnyvale
Next, I attempted to run the following command:
However, the process never completed, and the output seemed to stall. The output was:
Since this didn't work, I directly tried launching the perception module with the command:
Note that in my Apollo version, the /production/ folder mentioned in the guide doesn't exist. The output I got was:
The output contains many errors related to lane width. Skipping repetitive lines for brevity, the process eventually failed with:
I also tried with the next configuration:
setup-mode: Mkz Lgsvl
car: Kitti 140
map: Sunnyvale
Then run the next command:
And the output was:
And the last configuration I tried:
setup-mode: Mkz Lgsvl
car: Mkz Example
map: Sunnyvale
Then run the next command:
And the output was:
So, my questions would be the following:
Output from
apollo.sh config:Output from
cyber_launch start /apollo/modules/transform/launch/static_transform.launch:Output from
cyber_launch start /apollo/modules/perception/launch/perception_trafficlight.launchwith car Mkz Lgsvl 321:Output from
cyber_launch start /apollo/modules/perception/launch/perception_trafficlight.launchwith car Kitti 140:Output from
cyber_launch start /apollo/modules/perception/launch/perception_trafficlight.launchwith car Mkz Example:The text was updated successfully, but these errors were encountered: