search

tusen-ai / SST

Code for a series of work in LiDAR perception, including SST (CVPR 22), FSD (NeurIPS 22), FSD++ (TPAMI 23), FSDv2, and CTRL (ICCV 23, oral).
Apache License 2.0
801 stars 102 forks source link

What is the order to install the environment, if want to use CTRL #133

Closed atlantins closed 1 year ago

atlantins commented 1 year ago

i follow the order like this: 1.https://github.com/tusen-ai/SST/blob/main/docs/overall_instructions.md install pytorch mmcv mmdet mmsegmentation mmdet3d then pip install -e . 2.https://github.com/tusen-ai/SST/blob/main/docs/CTRL_instructions.md#step-2-use-immortaltracker-to-generate-tracking-results-in-training-split-bin-file-format (2)https://github.com/Abyssaledge/ImmortalTracker-for-CTRL run_mot.sh i really need a specific code running instruction to follow, and need enviroment installation order

Abyssaledge commented 1 year ago

The order you follow is right. Did any errors occur?

atlantins commented 1 year ago

1: can you give me a scrennshot in ./data, i am confused the structure, and https://github.com/open-mmlab/mmdetection3d/blob/1.0/docs/getting_started.md,the link you give for mm3d to getting started wasn't existed, So I don't know if the dataset I'm using in mmdet3d v1.1.0 is useful for CTRL i was confused about should we both use kiit_format and waymo_format,or just kitti_format?

2. $ python ./tools/ctrl/extract_poses.py Traceback (most recent call last): File "./tools/ctrl/extract_poses.py", line 31, in idx_str = info['point_cloud']['velodyne_path'].split('/')[-1].split('.')[0] TypeError: string indices must be integers

3. https://github.com/tusen-ai/SST/blob/main/docs/CTRL_instructions.md if i follow the instructions, should i use waymo_format data? my folder like this ./data/waymo_format (1)bash preparedata/waymo/waymo_preparedata.sh ./data/waymo (2)bash preparedata/waymo/waymo_convert_detection.sh ./data/waymo_format/gt.bin cp
or bash preparedata/waymo/waymo_convert_detection.sh ./data/kitti_format/train_gt.bin cp
(3) run_mot.sh

atlantins commented 1 year ago

when i run_mot.sh $ sh run_mot.sh Tracking time cost: 0.22504878044128418s evaluation/waymo/pred_bin.py:195: YAMLLoadWarning: calling yaml.load() without Loader=... is deprecated, as the default Loader is unsafe. Please read https://msg.pyyaml.org/load for full details. config = yaml.load(open(args.config_path, 'r')) Converting TYPE vehicle into WAYMO Format 0it [00:00, ?it/s] Merging different object types

Abyssaledge commented 1 year ago

No, you cannot use the dataset from mmdet3d v1.1.10, because the definition of box is different. Please use our code to generate the data. The point cloud data can be shared between different versions of mmdet3d (with minimal modifications), but the pickle info file needs to be regenerated.

Abyssaledge commented 1 year ago

If tracking time costs less than 1 second, there must be errors in the tracking process and the main body is skipped. And the errors would not be thrown due to the multi-process. So change it to single-process for debugging.

atlantins commented 1 year ago

so can i use python tools/create_data.py waymo --root-path --out-dir --workers 128 --extra-tag waymo in ./SST/ to generate data,or i should use mmdet3d v0.15.0 to generate data

Abyssaledge commented 1 year ago

Yes, you could use ‘tools/create_data’ to generate data. And our repo is exactly v0.15.0. Please reopen this issue if you need further discussion, or I may miss your posts due to too many issues.