Evaluation Visualization (2)

Excerpt from submitted algorithm results compared to WildDash's Ground Truth. Label images use the standard cityscapes color coding (white is used for unknown labels; e.g. ScanNet); difference images show correct pixels in green, pixels with incorrect class but correct category in yellow, and incorrect pixels in red. All shown frames are part of WildDash's public validation frames and have no direct impact on the leaderboard. However, the order in the list corresponds to the leaderboard order.

Frames visualized:

• py0006; Pouring rain in city scene from Paraguay
• hu0015; Overexposed frame near a Hungarian construction site
• ee0017; Dark night at a parking lot in Estonia

go back to leaderboard