add fast tracker and fix bugs of cbiou

Author: JackWoo0831

.gitignore

@@ -11,6 +11,7 @@ wandb/*
 track_result.txt
 .idea/
 track_results/*
+track_demo_results/*
 *.mp4
 *.mkv
 temp.py

README.md

@@ -24,6 +24,7 @@ However, bugs or issues should still be prioritized in the **Issue section in Gi
 
 ## 🗺️ Latest News
 
+- ***2025.11.28*** FastTracker is added. Fix lost tracklets bugs of CBIoU_tracker.
 - ***2025.7.8*** New version 2.1 released. Add ImproAssoc, TrackTrack and support TensorRT. The other details are as follows:
 
 <details>
@@ -66,6 +67,7 @@ and the tracker supports:
 - Hybrid SORT ([AAAI 2024](https://ojs.aaai.org/index.php/AAAI/article/view/28471))
 - ImproAssoc ([CVPRW 2023](https://openaccess.thecvf.com/content/CVPR2023W/E2EAD/papers/Stadler_An_Improved_Association_Pipeline_for_Multi-Person_Tracking_CVPRW_2023_paper.pdf))
 - TrackTrack ([CVPR 2025](https://openaccess.thecvf.com/content/CVPR2025/html/Shim_Focusing_on_Tracks_for_Online_Multi-Object_Tracking_CVPR_2025_paper.html))
+- FastTracker ([arxiv 2508](https://arxiv.org/pdf/2508.14370))
 
 and the reid model supports:
 
@@ -260,12 +262,21 @@ In addition, you can also specify
 
 - TrackTrack: `python tracker/track.py --dataset visdrone_part --detector yolo_ultra --tracker tracktrack --kalman_format bot --detector_model_path weights/yolov8l_VisDrone_35epochs_20230509.pt --save_images --nms_thresh 0.95 --reid`
 
+- FastTracker: `python tracker/track.py --dataset uavdt --detector yolo_ultra_v8 --tracker fasttrack --kalman_format byte --detector_model_path weights/yolov8l_UAVDT_60epochs_20230509.pt`
+
+
 > **Important notes for UCMC Track:**
 > 
 > 1. Camera parameters. The UCMC Track need the intrinsic and extrinsic parameter of camera. Please organize like the format of `tracker/cam_param_files/uavdt/M0101.txt`. One video sequence corresponds to one txt file. If you do not have the labelled parameters, you can refer to the estimating toolbox in original repo ([https://github.com/corfyi/UCMCTrack](https://github.com/corfyi/UCMCTrack)).
 > 
 > 2. The code does not contain the camera motion compensation part between every two frame, please refer to [https://github.com/corfyi/UCMCTrack/issues/12](https://github.com/corfyi/UCMCTrack/issues/12). From my perspective, since the algorithm name is 'uniform', the update of compensation between every two frames is not necessary.
 
+>**Important Notes on Fast Tracker**
+> 
+> In `fast_tracker.py`, the configuration related to the tracker is stored in the global variable `FAST_TRACKER_CONFIG`, which includes thresholds for recording occluded targets (such as velocity damping, bounding box enlargement, etc.) and environmental optimizations for road structure fusion (under the "ROIs" key, with specific values and meanings referenced in the original paper)
+
+
+
 ### ✨ TensorRT Convert and Inference
 
 This code supports **fully automatic** generation and reasoning of Tensor RT engine, **which can be used for both detection model and ReID model**. If you have not converted Tensor RT engine, just add `--trt` parameter when running, for example:

README_CN.md

@@ -18,6 +18,7 @@ git checkout v2.1  # change to v2.1 branch !!
 
 ## 🗺️ 最近更新
 
+- ***2025.11.28*** 增加FastTracker跟踪算法。修复CBIoU Tracker中的丢失轨迹的bug.
 - ***2025.7.8*** 新版本2.1发布. 添加ImproAssoc, TrackTrack并支持TensorRT. 其他细节如下:
 
 <details>
@@ -60,6 +61,7 @@ git checkout v2.1  # change to v2.1 branch !!
 - Hybrid SORT([AAAI 2024](https://ojs.aaai.org/index.php/AAAI/article/view/28471))
 - ImproAssoc ([CVPRW 2023](https://openaccess.thecvf.com/content/CVPR2023W/E2EAD/papers/Stadler_An_Improved_Association_Pipeline_for_Multi-Person_Tracking_CVPRW_2023_paper.pdf))
 - TrackTrack ([CVPR 2025](https://openaccess.thecvf.com/content/CVPR2025/html/Shim_Focusing_on_Tracks_for_Online_Multi-Object_Tracking_CVPR_2025_paper.html))
+- FastTracker ([arxiv 2508](https://arxiv.org/pdf/2508.14370))
 
 REID模型支持：
 
@@ -254,12 +256,18 @@ python tracker/track.py --dataset ${dataset name, related with the yaml file} --
 
 - TrackTrack: `python tracker/track.py --dataset visdrone_part --detector yolo_ultra --tracker tracktrack --kalman_format bot --detector_model_path weights/yolov8l_VisDrone_35epochs_20230509.pt --save_images --nms_thresh 0.95 --reid`
 
+- FastTracker: `python tracker/track.py --dataset uavdt --detector yolo_ultra_v8 --tracker fasttrack --kalman_format byte --detector_model_path weights/yolov8l_UAVDT_60epochs_20230509.pt`
+
 >**UCMC Track的重要提示：**
 > 
 > 1. 相机参数. UCMC Track需要相机的内参和外参. 请按照`tracker/cam_ram_files/uavdt/M0101.txt`的格式组织. 一个视频序列对应一个txt文件. 如果您没有标记的参数, 可以参考原始仓库中的估算工具箱([https://github.com/corfyi/UCMCTrack](https://github.com/corfyi/UCMCTrack)).
 > 
 > 2. 该代码不包含每两帧之间的相机运动补偿部分, 请参阅[https://github.com/corfyi/UCMCTrack/issues/12](https://github.com/corfyi/UCMCTrack/issues/12). 在我看来, 既然算法叫"统一相机运动补偿", 因此不需要每两帧之间再更新补偿. 
 
+>**Fast Tracker的重要提示：**
+> 
+> 在fast_tracker.py中，与跟踪器有关的配置在FAST_TRACKER_CONFIG全局变量中，包括对遮挡目标记录的相关阈值（速度阻尼、边界框放大等），以及融合道路结构的环境优化（"ROIs"键，具体数值以及含义请参照原论文）
+
 ### ✨ TensorRT的转换与推理
 
 该代码支持**全自动**的Tensor RT engine的生成与推理, **既可以用于检测模型, 也可以用于ReID模型**. 如果您还没有转换Tensor RT engine, 只需在运行时加上`--trt`参数, 例如:

tracker/track.py

@@ -32,6 +32,7 @@
 from trackers.hybridsort_tracker import HybridSortTracker
 from trackers.tracktrack_tracker import TrackTrackTracker
 from trackers.improassoc_tracker import ImproAssocTracker
+from trackers.fast_tracker import FastTracker
 
 # YOLOX modules
 try:
@@ -86,7 +87,8 @@
     'ucmctrack': UCMCTracker, 
     'hybridsort': HybridSortTracker, 
     'tracktrack': TrackTrackTracker, 
-    'improassoc': ImproAssocTracker
+    'improassoc': ImproAssocTracker, 
+    'fasttrack': FastTracker
 }
 
 def get_args():

tracker/track_demo.py

@@ -31,6 +31,7 @@
 from trackers.hybridsort_tracker import HybridSortTracker
 from trackers.tracktrack_tracker import TrackTrackTracker
 from trackers.improassoc_tracker import ImproAssocTracker
+from trackers.fast_tracker import FastTracker
 
 # YOLOX modules
 try:
@@ -84,7 +85,8 @@
     'ucmctrack': UCMCTracker, 
     'hybridsort': HybridSortTracker, 
     'tracktrack': TrackTrackTracker, 
-    'improassoc': ImproAssocTracker
+    'improassoc': ImproAssocTracker, 
+    'fasttrack': FastTracker, 
 }
 
 def get_args():
@@ -108,7 +110,7 @@ def get_args():
     parser.add_argument('--init_thresh', type=float, default=0.3, help='filter new detections, larger than this thresh consider as new tracklet')
     parser.add_argument('--nms_thresh', type=float, default=0.7, help='thresh for NMS')
 
-    parser.add_argument('--device', type=str, default='6', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
+    parser.add_argument('--device', type=str, default='0', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
 
     """yolox"""
     parser.add_argument('--num_classes', type=int, default=1)

tracker/trackers/byte_tracker.py

@@ -123,9 +123,12 @@ def update(self, output_results, img, ori_img):
                           (tlwh, s, cate) in zip(dets_second, scores_second, cates_second)]
         else:
             detections_second = []
+
         r_tracked_tracklets = [tracklet_pool[i] for i in u_track if tracklet_pool[i].state == TrackState.Tracked]
         dists = iou_distance(r_tracked_tracklets, detections_second)
+        
         matches, u_track, u_detection_second = linear_assignment(dists, thresh=0.5)
+        
         for itracked, idet in matches:
             track = r_tracked_tracklets[itracked]
             det = detections_second[idet]

tracker/trackers/c_biou_tracker.py

@@ -97,6 +97,13 @@ def update(self, output_results, img, ori_img):
                 track.re_activate(det, self.frame_id, new_id=False)
                 refind_tracklets.append(track)
 
+        # mark u_track as lost
+        for it in u_track:
+            track = unmatched_tracklets[it]
+            if not track.state == TrackState.Lost:
+                track.mark_lost()
+                lost_tracklets.append(track)
+                
         '''Deal with unconfirmed tracks, usually tracks with only one beginning frame'''
         detections = [unmatched_detections[i] for i in u_detection]
         dists = buffered_iou_distance(unconfirmed, detections, level=1)