The problem of predicting my own data set

[xiaochenwise]

I test my own dataset using your trained bike_human model, Enter python render_360.py --scene_dir data/mydata/output --weights_path out/bike_human/checkpoint.pth --mode canonical_360， we get the following result Excuse me, is there something wrong with me? Or would I have to retrain for my own data set

[jiangwei221]

Rednering canonical_360 only uses the scene scale to determine the interval compensation. It looks like there is some scale mismatch between mydata scene and bike scene. You need to retrain the scene model and human model using mydata.

[xiaochenwise]

Rednering canonical_360 仅使用场景比例来确定区间补偿。看起来mydata场景和bike场景之间存在一些比例不匹配。 您需要使用 重新训练场景模型和人体模型mydata。 Ok, I have another question. When you saved the output file of densepose in the preprocessing stage, did you save this data？