Please login before submit the results of your algorithm. Then check the verification email sent to you upon registration. If you have not received the verification email, please check your Junk-box in Email; if still not, please contact us.
The DETRAC-train dataset can be used to present results of your method with various settings (e.g. different parameters, features). The evaluation server for the DETRAC-test dataset is forbidden to be used for parameter tuning; otherwise the account may be permanently banned.
Each user can apply at most 2 different detectors/trackers/counters at a time. After 3 months, you can apply another new algorithm. You can only submit your results every 3 days.
We encourage that you can upload your source codes for contributing to the community.
Please submit your results as a single .zip file. Each video sequence corresponds to one .txt file. The detection result for each sequence must be stored in the archive's root folder. The file name must be exactly like the sequence name (case sensitive). For example:
./SequenceName1_Det_DetectorName.txt
./SequenceName2_Det_DetectorName.txt
./SequenceName3_Det_DetectorName.txt
...
In addition, the detection result should be stored in CSV (Comma-Separated Values) style, where each value is separated by a comma. In every detection file, the bounding box positions and corresponding normalized scores (with the range [0,1]) should be provided. Moreover, each line of the file should be set as Frame, Number, Left, Top, Width, Height, Score, Type and Color in order.
Here is an example of the detection file:
1,1,752.59,352.17,156.25,116.46,0.48289,1,1 2,1,745.75,345.89,150.74,111.73,0.46823,1,2 2,2,298.26,205.98,61.21,30.37,0.040935,1,3 3,1,736.32,337.48,149.25,107.83,0.47122,2,4 3,2,613.37,65.793,28.16,43.027,0.040239,3,4 4,1,730.13,336.44,142.06,101.93,0.42813,4,1 4,2,158,206.58,33.33,26.61,0.023112,2,1 ...
TIPS: The detection results obtained by your detector should be organized in the above format. It is worth mentioning that the last two columns (vehicle type and color) are optional for the AVSS 2019 Challenge based on the dataset with the V3 verson annotation. The index for 6 main vehicle types: 1 (Sedan), 2 (SUV), 3 (Van), 4 (Taxi), 5 (Bus), and 6 (Truck) and the index for 10 main vehicle colors: 1 (White), 2 (Silver), 3 (Black), 4 (Gray), 5 (Blue), 6 (Red), 7 (Brown), 8 (Green), 9 (Yellow) and 10 (Multi). To train the detector, the samples from other datasets are accepted.
Please submit your results as a single .zip file. Each video sequence should have five .txt files. The Speed file records the running speed of the tracker. We use frame-Per-second (fps) to measure the speed of the tracker. Four files with the position suffix (i.e., LX, LY, W and H) are the position result, which stand for Left, Top, Width, Height information of the bounding box on the object trajectories generated by the tracker. The results for each sequence must be stored in the archive's root folder. The file name must be exactly like the sequence name (case sensitive) with the corresponding suffix. For example:
./SequenceName1_LX.txt
./SequenceName1_LY.txt
./SequenceName1_W.txt
./SequenceName1_H.txt
./SequenceName1_Speed.txt
./SequenceName2_LX.txt
./SequenceName2_LY.txt
./SequenceName2_W.txt
./SequenceName2_H.txt
./SequenceName2_Speed.txt
...
Specifically, the position result should be stored in CSV (Comma-Separated Values) style, where each value is separated by a comma. In each .txt file, the row represents the frame number and the column represents the ID of the target. For example, the j-th target shows up in the i-th frame, and the corresponding element should be the position value such as left position, top position, width and height. Note that the position value is counted starting from 1. If no target appears, the value should be 0. Here is an example of the height file:
84.88, 0, 0, 72.75, ... 90.98, 0, 55.14, 63.33, ... 84.88, 0, 55.14, 67.88, ... 90.97, 0, 48, 72.76, ... 84.88, 0, 48, 72.76, ... 84.88, 0, 0, 63.34, ... ...
TIPS: The tracking results can be obtained by the DETRAC-MOT toolkit. An example submission package (i.e., the results of the tracker GOG with the detector DPM) for tracking is provided here for better understanding. You can choose one provided detector (i.e., DPM, ACF, R-CNN, and CompACT) or your own detector to perform tracking. If you choose your own detector, please present the methodology of the detector (with its name) in the submitted document. We encourage you make the input detections publicly available, or private detection will be displayed as "N/A" (i.e. not available) in the submitted result page.
Please submit your results as a single .zip file. Each video sequence corresponds to one .txt file. The counting result for each sequence must be stored in the archive's root folder. The file name must be exactly like the sequence name (case sensitive). For example:
./SequenceName1_Cnt_CounterName.txt
./SequenceName2_Cnt_CounterName.txt
./SequenceName3_Cnt_CounterName.txt
...
In addition, the detection result should be stored in CSV (Comma-Separated Values) style, where each value is separated by a comma. In every counting file, the counts of vehicles and the corresponding type and color should be provided. Moreover, each line of the file should be set as Count, Vehicle Type, Vehicle Color in order. Here is an example of the counting file:
12,1,1 5,1,2 2,2,3 0,3,2 ...
TIPS: The counting results obtained by your algorithm should be organized in the above format. The counting task is conducted for the AVSS 2019 Challenge based on the dataset with the V3 verson annotation. The index for 6 main vehicle types: 1 (Sedan), 2 (SUV), 3 (Van), 4 (Taxi), 5 (Bus), and 6 (Truck) and the index for 10 main vehicle colors: 1 (White), 2 (Silver), 3 (Black), 4 (Gray), 5 (Blue), 6 (Red), 7 (Brown), 8 (Green), 9 (Yellow) and 10 (Multi).