zqzqz / advtrajectoryprediction Goto Github PK

I wanted to know if you have any scripts or formulae to calculate the bounds, xy_distribution, detect_opts values for the datasets as done in AdvTrajPrediction/prediction/dataset/apolloscape.py and other dataset specific files in dataset folder.

It will be great if you can share this information as it will help us for testing other datasets as well.

Hi there, thank you for releasing the codes. When trying to reproduce the results by the instructions, I met the error "No such file or directory: '/home/xxxx/AdvTrajectoryPrediction/prediction/dataset/../../data/nuScenes/map_name.txt' ". It was thrown out when I ran the test.py or generate_data.py. I was wondering whether this file is missing or I need to generate it. Thanks!

After running the code, I found that the error in the attack results is larger than that in the paper. Specifically, under single frame attack conditions, the results of the model that was not attacked were consistent with the results in the paper, but the results after using a single frame attack were much better than those in the paper, such as fqa apollscape. The results I ran were about 90% higher than those in the paper

Thanks

Hi Qingzhao, thank you for releasing the code with us!

After following the steps on the readme file, I tested the normal trajectron_map model on nuScenes dataset with single_frame mode. The ADE and FDE I calculated are 7.131660 and 17.831819, but the results on paper are 2.69 and 5.39. I'm wondering if I miss some important steps to reproduce the result or if I need to modify anything locally?

Thanks!

Hi, in the requirement.txt
there are a few packages using the local file.

dataclasses @ file:///tmp/build/80754af9/dataclasses_1614363715916/work

which emit the error "file not found" while doing pip install.

Hi
I cloned the AdvTrajectoryPrediction repo and completed the set up following the instructions. I am running the test.py on a high performance compute cluster with access to a Tesla V100 GPU but noticed that multi_frame mode tests do not complete even after 12 hours. When I check GPU utilization it's in the range of 0-7%.

I know that this code was originally written for an RTX 2080 with CUDA 10.2 and the Trajectron model that I am using for testing also optimized their code for GPU but I am unable to locate the source of delay even after going through all the files.

Is it supposed to take this long or am I missing something? How can I make the testing faster?

Below is the command I am running:

python test.py --dataset apolloscape --model trajectron --mode single_frame --overwrite

Thanks in advance!