Extract features and evaluation parameters with OpenCV in PolyLaneNet

yutaekoh / LaneLineDetection

5 stars 0 forks source link

Extract features and evaluation parameters with OpenCV in PolyLaneNet #9

Open juuwong opened 3 years ago

juuwong commented 3 years ago

Brainstorm:

extract basic evaluation parameters provided by TuSimple dataset
implement confusion matrix to visualize false negative and false positive rates
comparison between road conditions with TuSimple dataset (ie. sunny/cloudy; straight, curved; road construction; lane patterns)
note potential areas of improvement based on this data

juuwong commented 3 years ago

Unable to generate full video yet. Researching how to generate .pkl predictions file to run gen_video.py file in PolyLaneNet

PolyLaneNet/utils/gen_video.py / line 12 parser.add_argument("--pred", help=".pkl file to load predictions from")

No .pkl file is generated after testing the model yet.

juuwong commented 3 years ago

Encountered an issue with gen_video.py (base) root@462edb4c0a0b:/workspace# python gen_video.py --pred experiments/tusimple/models/model_2695.pt --cfg cfgs/modified_atlas_tusimple_fulltrain.yaml total annos 2782 Transforming annotations... Done. Using resolution 640x360 Traceback (most recent call last): File "gen_video.py", line 67, in main() File "gen_video.py", line 46, in main for idx, pred in tqdm(zip(range(len(dataset)), predictions), total=len(dataset)): TypeError: 'int' object is not iterable

juuwong commented 3 years ago

Unable to generate full video yet. Researching how to generate .pkl predictions file to run gen_video.py file in PolyLaneNet

PolyLaneNet/utils/gen_video.py / line 12 parser.add_argument("--pred", help=".pkl file to load predictions from")

No .pkl file is generated after testing the model yet.

The .pkl file can be substituted with the pretrained model file (model_2695.pt)

juuwong commented 3 years ago

Brainstorm:

extract basic evaluation parameters provided by TuSimple dataset

implement confusion matrix to visualize false negative and false positive rates

comparison between road conditions with TuSimple dataset (ie. sunny/cloudy; straight, curved; road construction; lane patterns)

note potential areas of improvement based on this data

Unable to connect to X server issue to complicated to resolve when running from docker container over SSH. Will attempt running gen_video.py to output a video file of predictions