Inference on Single Image

[joeyism]

Hi guys, this is great work. I was wondering if how I can run inference just on a single image, as a test on how well the model generates. For example, if I provide just an image of the couch (as seem in the demo), what do I need to execute to be able to generate the mesh?

[DaLi-Jack]

Hi! This is a good question.
In our method, mesh recovery from a single image is divided into two stages, the first stage is single-image 3D object detection, and another stage is 3D object reconstruction which relies on the object detection results. We follow InstPIFu to use the Im3D detector as the single-image 3D object detector, for more details please refer to #2. This repo only releases the 3D object reconstruction codes, so if you want to infer on a single image, please refer to InstPIFu Object Detetion to get the 3D object detection results, and then use SUNRGBD code inference_sunrgbd.py for the 3D object reconstruction.