search

voldemortX / pytorch-auto-drive

PytorchAutoDrive: Segmentation models (ERFNet, ENet, DeepLab, FCN...) and Lane detection models (SCNN, RESA, LSTR, LaneATT, BézierLaneNet...) based on PyTorch with fast training, visualization, benchmarking & deployment help
BSD 3-Clause "New" or "Revised" License
824 stars 137 forks source link

Testing the LaneDetection on CULane dataset without training #130

Closed aghari19 closed 5 months ago

aghari19 commented 1 year ago

Hello,

I would like to just test the CULane dataset with the curve modeling algorithm to see how it performs. Do I have to train before I run the --test?

Please let me know, because I am doing a comparison for a class project and do not have access to GPUs or CUDA library.

Thanks, Haritha

voldemortX commented 1 year ago

@aghari19 You can just download pre-trained weights in MODEL_ZOO. There is no need for training.

aghari19 commented 1 year ago

@voldemortX Thank you for letting me know about it. I have downloaded the pre-trained weights however, I am having trouble with running the algorithm with it. Could you let me know how to run with the pre-trained weights

Update: I am sorry for the trouble. I was able to figure it out and got it running. Thank you for all the help :)

aghari19 commented 1 year ago

I have a follow up question: I was able to successfully able to run this command python main_landet.py --test --config=configs/lane_detection/baseline/vgg16_culane.py --mixed-precision

At the end it said it created the 'tb_logs' file. However, when I grep for this file it is not found. I am not sure what the output of this command would be because I would like to visualize after the images with the lanes detected. So I was thinking this command would take in a folder of images and output various .txt files with the lanes corresponding to each image.

Then I could use the visualizer to see the lanes detected. Is my understanding incorrect?

Please let me know, Thanks

voldemortX commented 1 year ago

@aghari19 the tb_logs are tensorboard logs mainly used for training. The predicted results are in ./output.

You could also checkout our visualization tools in https://github.com/voldemortX/pytorch-auto-drive/blob/master/docs/VISUALIZATION.md, they can include inference and vis on image folder or videos. Most datasets are supported as well, but in the advanced tutorial (you need to modify the config for this).

voldemortX commented 5 months ago

The problem seems to be solved. Feel free to comment or open new issues.