shunsukesaito / PIFu

I have see that in paper you are trying to convert img to 3d model.
So the img should be real img (camera clicked)

SO why is Pifu using a rendered img for training?

Line 193 in 02a6d86

    
           rndr_uv.set_mesh(vertices, faces, normals, faces_normals, textures, face_textures, prt, face_prt, tan, bitan)

PIFu/apps/render_data.py

Line 194 in 02a6d86

rndr_uv.set_albedo(texture_image)

PIFu/apps/render_data.py

Line 243 in 02a6d86

    
           cv2.imwrite(os.path.join(out_path, 'RENDER', subject_name, '%d_%d_%02d.jpg'%(y,p,j)),255.0*out_all_f)

Why dont we just train using the original jpg img? Can't we just resize it ?

Well. PIFu is trained on textured meshes, so you need to render them somehow. If of course you do have the raw data used for generating those you could train on the actual images.

Why Use render img over original image?