Upload Your Results

To submit your results, please upload a json file by clicking the button below. The file should contain the results of your model on the CADOT.

Upload Results

Our Baseline Performance

To establish a clear benchmark for participants, we defined baseline performance using four state-of-the-art object detection models: YOLOv11, YOLOv12, Faster R-CNN, and DiffusionDet. For each model, we detail the experimental settings and configurations that were used to obtain the reported results, ensuring transparency and reproducibility.

YOLOv11/YOLOv12

For YOLOv11 and YOLOv12, we adopted the default training settings as specified in their respective publications. Both models are trained using a learning rate of 0.01 with the SGD optimizer, incorporating a momentum of 0.937 and a weight decay of 0.0005 to mitigate overfitting. Training is conducted over 300 epochs with a batch size of 16.

Faster R-CNN

we used the official implementation from Faster R-CNN paper using a VGG-16 backbone. The model is trained with the SGD optimizer, a learning rate of 0.005, a momentum of 0.9, and a weight decay of 0.0001. To improve robustness, we incorporate multi-scale training and applied standard data augmentations such as horizontal flipping and random cropping. All remaining hyperparameters follow the default settings provided in the Fast R-CNN paper.

DiffusionDet

For DiffusionDet, we used the Swin Transformer as the backbone architecture. The model is optimized using AdamW with a base learning rate of 2.5×10⁻⁵ and a weight decay of 1×10⁻⁴. Training is performed over 450,000 iterations. We applied the default data augmentation strategies, including RandomFlip, RandomResizedCrop, and RandomCrop.