pmh47 / dirt

DIRT: a fast differentiable renderer for TensorFlow
MIT License
313 stars 63 forks source link

The paper in IJCV shows the rendering detials, but I also have some question about it's process #77

Closed ljx970128 closed 4 years ago

ljx970128 commented 4 years ago

When use the subdivision method, the decoder net will output the mesh parameters include vertex offsets. I want to know the specific format of these vertex offsets, whether to include only the coordinates of surface points(98 vertices, 294 parameters). The faces in "sample" are pre-defined, so in your paper's experiment these faces how to get? The decoder net would output rotation parameter, what's different between it and the encoder's output theta, the pose parameter. @pmh47 hope to receive your reply.

pmh47 commented 4 years ago

Closing as this is not related to DIRT itself (I replied to you by email re the IJCV paper).