DMKPD
DM Code Key Point Detection Algorithm Based on CenterNet
- The reading of DM code usually includes positioning and decoding. Accurate positioning is a prerequisite for successful decoding. Further image processing methods are required for precise positioning, which will reduce efficiency.
- Because of the above problems, a CenterNet-based DM code key point detection network is proposed, which can directly obtain the four key points of the DM code. Compared with the existing methods, the degree of fitness is higher, which is conducive to direct decoding. To further improve the positioning accuracy, an enhanced loss function is designed, including DM code key point heatmap loss, standard DM code projection loss, and polygon Intersection-over-Union (IoU) loss, which is beneficial for the network to learn the spatial geometric characteristics of DM code.
Submission Guidelines
DM Keypoint Detection Benchmark
1. What to Submit
Participants must submit a ZIP file containing exactly one prediction txt file for each of the 300 evaluated images. The benchmark uses a single class: DPM.
- Each evaluated image must correspond to exactly one
.txtfile. - Filename must match the image basename.
- If an image has no detected target, submit an empty txt file.
- Missing files and extra files will cause the submission to be rejected.
2. File Naming Rules
Prediction files must follow this naming rule:
{image_basename}.txt
Examples:
adv_final_02074.png→adv_final_02074.txtdm0077.jpg→dm0077.txt
The filename extension of the image is not included in the txt filename. The basename must match exactly.
3. Per-line Prediction Format
Each line in a txt file describes one detected object:
DPM conf x1 y1 x2 y2 x3 y3 x4 y4
A single image may contain multiple targets, so one txt file may contain multiple lines. Coordinates are the four corner points of the quadrilateral target.
Please submit the corner points in this order: top-left → top-right → bottom-right → bottom-left. If the order is incorrect, the evaluation script will normalize the point order before evaluation.
Coordinates must use the original image pixel coordinate system. Do not normalize coordinates to the range [0, 1].
| Field | Type | Description |
|---|---|---|
| 1 | string | Class name, fixed value: DPM |
| 2 | float | Confidence score in [0, 1] |
| 3-10 | float |
Four corner coordinates:
x1 y1 x2 y2 x3 y3 x4 y4.
The recommended order is top-left, top-right, bottom-right, bottom-left.
|
Predictions with confidence scores below 0.3 will not participate in evaluation.
4. Submission Example
Example content in one prediction txt file (assuming there are three objects in the image):
DPM 0.9275 574 377 751 362 755 555 580 570 DPM 0.9253 783 599 903 593 908 715 784 721 DPM 0.9013 561 772 736 767 742 957 566 960
If no target is detected in one image, keep the corresponding txt file empty.
5. ZIP Folder Structure
Your final upload must be a ZIP file with txt files organized like this:
map_out/ ├── 00012.txt ├── 00018.txt ├── 00020.txt └── ...
The ZIP file must contain one txt file for every evaluated image and must not contain any extra files.
6. Metrics
The leaderboard displays the following keypoint detection metrics:
- AP_50_95
- AP_50
- AP_75
- AR_50_95
OKS evaluation uses four equal keypoint sigmas of 0.1.
Other metrics are not displayed on the leaderboard.
7. Important Notes
Do not change the class name. The class must always be DPM.
Do not submit extra files. Only the required txt prediction files should be included.
Do not omit empty txt files. Images without detections must still have an empty txt file.
Upload a ZIP file only.
File size limit: your submission ZIP must be within the platform upload limit.