The More You See in 2D, the More You Perceive in 3D

Supplementary Material

 

• More Sight - More Insight (Section 4.2)
1. 3D reconstruction
2. Novel View Synthesis (with and without finetuning)
• System Verification - 3D reconstruction (Section 4.2)

 

---

Click Play/Pause All on the bottom-right to play and pause all the videos.

---

More Sight - More Insight

---

1. 3D Reconsturction

We present the 3D reconstructions for a varying number (1 to 5) of input images to SAP3D.

As shown, with more input images, the 3D reconstruction of SAP3D improves.

---

One input (Zere-1-to-3)	Two inputs	Three inputs	Four inputs	Five inputs
	+	+	+	+
Results:

---

One input (Zere-1-to-3)	Two inputs	Three inputs	Four inputs	Five inputs
	+	+	+	+
Results:

---

One input (Zere-1-to-3)	Two inputs	Three inputs	Four inputs	Five inputs
	+	+	+	+
Results:

---

One input (Zere-1-to-3)	Two inputs	Three inputs	Four inputs	Five inputs
	+	+	+	+
Results:

---

One input (Zere-1-to-3)	Two inputs	Three inputs	Four inputs	Five inputs
	+	+	+	+
Results:

---

One input (Zere-1-to-3)	Two inputs	Three inputs	Four inputs	Five inputs
	+	+	+	+
Results:

---

One input (Zere-1-to-3)	Two inputs	Three inputs	Four inputs	Five inputs
	+	+	+	+
Results:

2. Novel View Synthesis

We present generated novel views for a varying number (1 to 5) of input images to SAP3D.

For each input set, we provide the novel views before and after the fine-tuning stage. We use stochastic conditioning in all the experiments.

Similarily to 3D reconstruction, with more input images, the novel view synthesis of SAP3D improves. Moreover, the adapation stage improves the consistency of the generations.

---

One input (Zere-1-to-3)	Two inputs	Three inputs	Four inputs	Five inputs
	+	+	+	+
SAP3D without adaptation:
SAP3D:

---

One input (Zere-1-to-3)	Two inputs	Three inputs	Four inputs	Five inputs
	+	+	+	+
SAP3D without adaptation:
SAP3D:

---

One input (Zere-1-to-3)	Two inputs	Three inputs	Four inputs	Five inputs
	+	+	+	+
SAP3D without adaptation:
SAP3D:

---

One input (Zere-1-to-3)	Two inputs	Three inputs	Four inputs	Five inputs
	+	+	+	+
SAP3D without adaptation:
SAP3D:

---

One input (Zere-1-to-3)	Two inputs	Three inputs	Four inputs	Five inputs
	+	+	+	+
SAP3D without adaptation:
SAP3D:

---

System Verification on 3D Reconstruction (Section 4.3)

We evaluate the effects of:

1. Initial camera pose estimation: RelPose++ compared to our scaled version - "SAP3D without adaption"

2. Fine-tuning stage: SAP3D without adaption compared to SAP3D.

We compare performmance of 3D reconstruction and novel view synthesis, when 3 input images are provided.

---

	Input images:
RelPose++	SAP3D without adapation	SAP3D

---

	Input images:
RelPose++	SAP3D without adapation	SAP3D