2025-03-09
SDTrack preprint is now available on arXiv.
2025-03-15
SDTrack code is now available.
2025-08-14
🏆 🏆 🏆 We extended SDTrack to spiking cameras, introduced three datasets generated using spiking-camera simulators, and provided the first tracking baseline for spiking cameras. This work received the Best Dataset & Benchmark Award (sole winner) in the IJCAI 2025 SpikeCV–Wuji Challenge (Dataset & Benchmark Track), along with a $1,500 prize. We thank the SpikeCV organizing committee. Link
2025-09-19
At NeurIPS 2026, 4 accepted papers (NeurIPS-1, NeurIPS-2, NeurIPS-3, NeurIPS-4) built upon our model or adopted our reported baseline. Congratulations to the authors, and thank you for your interest!
2025-12-19
A neuromorphic tracking system built upon the DAVIS346 camera, the SDTrack pipeline, and Lynxi chips was showcased at the SLAI (Shenzhen Loop Area Institute) – LIMA (Language, Intelligence and Machines) Center Open Day. Link
2025-12-28
The same neuromorphic tracking system was showcased at the 5th Brain Science Frontier and Industry Conference and 2025 Shenzhen Brain-Computer Interface and Human-Machine Interaction Technology Expo.
2026-02-21
SDTrack has been accepted by CVPR 2026.
For the event-based camera component, please refer to SDTrack-Event.
For the spike camera component, please refer to SDTrack-Spike.
If you use our tracker(SDTrack), Event Aggregation method(GTP), reported experimental results, or any other original content from this work, please cite our paper:
@article{shan2025sdtrack,
title={Sdtrack: A baseline for event-based tracking via spiking neural networks},
author={Shan, Yimeng and Ren, Zhenbang and Wu, Haodi and Wei, Wenjie and Zhu, Rui-Jie and Wang, Shuai and Zhang, Dehao and Xiao, Yichen and Zhang, Jieyuan and Shi, Kexin and others},
journal={arXiv preprint arXiv:2503.08703},
year={2025}
}
Change the code in 'SDTrack/lib/train/run_training.py'
parser.add_argument('--local_rank', default=-1, type=int, help='node rank for distributed training')to
parser.add_argument('--local-rank', default=-1, type=int, help='node rank for distributed training')Change the code in 'SDTrack/lib/train/data/loader.py'
from torch._six import string_classesto
string_classes = str/xxx/SDTrack/lib/train/../../lib/train/data/loader.py:88: UserWarning: An output with one or more elements was resized since it had shape [3145728], which does not match the required output shape [1, 16, 3, 256, 256]. This behavior is deprecated, and in a future PyTorch release outputs will not be resized unless they have zero elements. You can explicitly reuse an out tensor t by resizing it, inplace, to zero elements with t.resize_(0). (Triggered internally at /opt/conda/conda-bld/pytorch_1720538622298/work/aten/src/ATen/native/Resize.cpp:28.)
Add the code in the top of '/SDTrack/lib/train/run_training.py'
import warnings
warnings.filterwarnings("ignore", category=UserWarning)[rank0]: (1) In PyTorch 2.6, we changed the default value of the `weights_only` argument in `torch.load` from `False` to `True`. Re-running `torch.load` with `weights_only` set to `False` will likely succeed, but it can result in arbitrary code execution. Do it only if you got the file from a trusted source.
[rank0]: (2) Alternatively, to load with `weights_only=True` please check the recommended steps in the following error message.
[rank0]: WeightsUnpickler error: Unsupported global: GLOBAL argparse.Namespace was not an allowed global by default. Please use `torch.serialization.add_safe_globals([argparse.Namespace])` or the `torch.serialization.safe_globals([argparse.Namespace])` context manager to allowlist this global if you trust this class/function.Solution: Add the parameter weights_only=False to all torch.load functions.
When using an environment similar to that of the 4090 on H100 GPU for training, there is an error. It seems to be due to an incompatibility between PyTorch version and H100 GPU.
Traceback (most recent call last):
File "lib/train/run_training.py", line 110, in <module>
main()
File "lib/train/run_training.py", line 102, in main
run_training(args.script, args.config, cudnn_benchmark=args.cudnn_benchmark,
File "lib/train/run_training.py", line 76, in run_training
expr_func(settings)
File "/data/xxx/xxx/SDTrack/lib/train/../../lib/train/train_script.py", line 63, in run
net = DDP(net, device_ids=[settings.local_rank], find_unused_parameters=True)
File "/data/xxx/xxx/anaconda3/envs/SDTrack-old/lib/python3.8/site-packages/torch/nn/parallel/distributed.py", line 580, in __init__
self._sync_params_and_buffers(authoritative_rank=0)
File "/data/xxx/xxx/anaconda3/envs/SDTrack-old/lib/python3.8/site-packages/torch/nn/parallel/distributed.py", line 597, in _sync_params_and_buffers
self._distributed_broadcast_coalesced(
File "/data/xxx/xxx/anaconda3/envs/SDTrack-old/lib/python3.8/site-packages/torch/nn/parallel/distributed.py", line 1334, in _distributed_broadcast_coalesced
dist._broadcast_coalesced(
RuntimeError: CUDA error: invalid device function
CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.Additionally, after simple testing, it was found that installing PyTorch on A800 server with the environment provided by this project results in errors, which are suspected to be related to specific configurations and versions.
The create_SDTrack_env.sh script provided in this project is derived from the actual running environment. If you encounter any environmental issues, please feel free to raise an issue, and I will assist in resolving them.