Novel View Synthesis from only a 6-DoF Camera Pose by Two-stage Networks

extreme simple idea

• pose -> coarse image(conditioned on training image embedding) : L1loss
• refine image : L1, perceptual, adversarial