Cityscapes Dataset and SSD Object Detection

[maryam-mohebbi]

Implement SSD object detection using the Cityscapes dataset.

• Familiarize with and preprocess the Cityscapes dataset for object detection.
• Implement and train the SSD (Single Shot MultiBox Detector) model for object detection.
• Test and optimize the SSD model for urban environment object detection using Cityscapes.

[maryam-mohebbi]

Related links: https://arxiv.org/pdf/1508.06576.pdf https://arxiv.org/pdf/1605.04603.pdf https://arxiv.org/abs/1512.02325#:~:text=,a%20single%20deep%20neural%20network https://arxiv.org/abs/1604.01685#:~:text=Cityscapes%20is%20comprised%20of%20a,labeled%20data

[maryam-mohebbi]

Implementation Steps:

• [ ] Load and preprocess the image data.
• [ ] Parse the label data to understand the annotation format.
• [ ] Preprocess the label data to match the format required by the SSD model.
• [ ] Set up the SSD model for training with KITTI data.
• [ ] Train the model.
• [ ] Evaluate the model on the testing data.

[maryam-mohebbi]

Lable guild for Kitti dataset: Example: Pedestrian 0.00 0 -0.20 712.40 143.00 810.73 307.92 1.89 0.48 1.20 1.84 1.47 8.41 0.01

• Object Class: "Pedestrian". This is the class label of the object.
• Truncation: "0.00". This indicates how much of the object is out of view. A value of 0 means fully visible, while 1 would mean completely truncated (out of view).
• Occlusion: "0". This represents the degree of occlusion. Generally, it's an integer (0: fully visible, 1: partly occluded, 2: largely occluded, 3: unknown).
• Observation Angle: "-0.20". This is the observation angle of the object, typically in radians or degrees.
• Bounding Box: "712.40 143.00 810.73 307.92". These are the pixel coordinates of the bounding box in the image. It's formatted as [xmin, ymin, xmax, ymax].
• 3D Object Dimensions: "1.89 0.48 1.20". These are the dimensions of the object (height, width, length) in meters.
• 3D Object Location: "1.84 1.47 8.41". This is the 3D location of the object relative to the camera, in meters.
• Rotation Y: "0.01". This is the rotation of the object around the Y-axis, often in radians.