BlockGAN – Supplemental material and results

BlockGAN: Learning 3D Object-aware Scene Representations from Unlabelled Images
Thu Nguyen-Phuoc, Christian Richardt, Long Mai, Yong-Liang Yang and Niloy Mitra
NeurIPS 2020

➤ Supplemental video (MP4)
➤ Supplemental document (PDF)

➤ Results on the CLEVRn datasets (n = 2, 3, 4)
➤ Results on the Synth-Carn datasets (n = 1, 2, 3)
➤ Results on the Synth-Chairn datasets (n = 1, 2, 3)
➤ Results on the natural Real-Cars dataset

Results on the CLEVRn datasets (n = 2, 3, 4)   🔝

Rotating objects

Horizontal translation

Translation along depth

Editing the background with fixed foreground (varying z0 with fixed z1)

Editing the foreground with fixed background (varying z1 with fixed z0)

Results on the Synth-Carn datasets (n = 1, 2, 3)   🔝

Rotating objects

Horizontal translation

Translation along depth

Editing the background with fixed foreground (varying z0 with fixed z1)

Editing the foreground with fixed background (varying z1 with fixed z0)

Results on the Synth-Chairn datasets (n = 1, 2, 3)   🔝

Rotating objects

Horizontal translation

Translation along depth

Editing the background with fixed foreground (varying z0 with fixed z1)

Editing the foreground with fixed background (varying z1 with fixed z0)

Results on the natural Real-Cars dataset   🔝

Rotating objects

Horizontal translation

Editing the background with fixed foreground (varying z0 with fixed z1)

Editing the foreground with fixed background (varying z1 with fixed z0)

Adding objects (Real-Cars dataset)

Stretching objects (Real-Cars dataset)

Slicing objects (Real-Cars dataset)