search

qianguih / voxelnet

This is an unofficial inplementation of VoxelNet in TensorFlow.
649 stars 240 forks source link

Problem with pre-trained model #18

Open ThorsteinnJonsson opened 6 years ago

ThorsteinnJonsson commented 6 years ago

There seems to be an issue with the pre-trained model included (save_model/pre_trained_car). When I run test.py, the visulizations seem to give highly inaccurate results. Furthermore, the ground truth boxes also seem to be flipped along the y-axis in the top-down lidar visualizations. Is there something wrong with the pre-trained model or coordinate conversions? Below are two example results.

000001_bv 000001_front

000002_bv 000002_front

ArpitaSTugave commented 6 years ago

I got the correct predictions. My guess would be your calibration values are wrong. Please update according to calibrated (synced) kitti data.

sakshi404 commented 6 years ago

Got the error while running pre-trained model: FailedPreconditionError : Attempting to use uninitialized value MiddleAndRPN_/conv20/bias was able to run it properly in next iteration, but again getting the same error. con12/beta or conv2/bias uninitialized .

What should be the value of available_gpu parameter if we are running on CPU? Do we have to change other things if model is run on cpu.

sainisanjay commented 5 years ago

@sakshi404, i am also getting same error, did you solve the above error????

sainisanjay commented 5 years ago

@ThorsteinnJonsson @qianguih may i know how to visualise the results. Do i have to pass any flag?? I am running the test script: python3 test.py -n pre_trained_car using this it save the result in ./predictions/data in .txt files.

angiend commented 5 years ago

@ThorsteinnJonsson @qianguih @sainisanjay thanks,Would you share the method to visualize the results?appreciate it

angiend commented 5 years ago

There seems to be an issue with the pre-trained model included (save_model/pre_trained_car). When I run test.py, the visulizations seem to give highly inaccurate results. Furthermore, the ground truth boxes also seem to be flipped along the y-axis in the top-down lidar visualizations. Is there something wrong with the pre-trained model or coordinate conversions? Below are two example results.

000001_bv 000001_front

000002_bv 000002_front how to visiualise the results?thanks

gkadusumilli commented 4 years ago

@ThorsteinnJonsson @qianguih may i know how to visualise the results. Do i have to pass any flag?? I am running the test script: python3 test.py -n pre_trained_car using this it save the result in ./predictions/data in .txt files.

for visualization run python3 test.py -n pre_trained_car --vis==True

SamudhbhavPrabhu commented 4 years ago

@ThorsteinnJonsson Even I am getting the same results. Were you able to solve this issue? @ArpitaSTugave Could you please post how your results look like for the first 2 images (for the same ones posted above by @angiend and @ThorsteinnJonsson )?