If a person is trained on a flat ground, the model's predictions for detecting people on sloped surfaces may be inaccurate. However, if the sloped surface is corrected using an algorithm, the predictions for people on top of it become accurate again. Additionally, it seems that for the same object, discrepancies between the height during training and the actual height during detection can also lead to inaccuracies in the predicted bounding boxes.
fffyyy111
commented
4 weeks ago
If a person is trained on a flat ground, the model's predictions for detecting people on sloped surfaces may be inaccurate. However, if the sloped surface is corrected using an algorithm, the predictions for people on top of it become accurate again. Additionally, it seems that for the same object, discrepancies between the height during training and the actual height during detection can also lead to inaccuracies in the predicted bounding boxes.