Associating Objects and Their Effects in Video

 Through Coordination Games

Supplementary Material

Main Project Page

 

• Supplementary PDF - with formulation and implementation details


• Results on 4-Object Reflections (Section 5.3, Table 1)
• Results on 2-Object Reflections (Section 5.3, Table 1)


• Results on 4-Object Shadows (Section 5.3, Table 1)
• Results on 2-Object Shadows (Section 5.3, Table 1)


• Results on Real Videos (Section 5.5)

 

---

Results on 4-Object Reflections

We demonstrate results of our model trained on the 2-Object Reflections dataset and tested on the 4-Object Reflections dataset (consisting of held-out objects and backgrounds). We show comparisons with Lu et. al [1].

0

1

2

3

More Reflection video test results here: 4 objects, 2 objects

 

---

Results on 4-Object Shadows

We demonstrate results of our model trained on the 2-Object Shadows dataset and tested on the 4-Object Shadows dataset (consisting of held-out objects and backgrounds). We show comparisons with Lu et. al [1].

0

1

2

3

More Shadow video test results here: 4 objects, 2 objects

 

---

Results on Real Videos

We demonstrate results on adapting our synthetic pretrained model to real videos and show comparisons with Lu et. al [1].

"Trampoline"

"Soccer"

"Dogwalk"

"Reflections"

 

---

References

[1] Erika Lu, Forrester Cole, Tali Dekel, Andrew Zisserman, William T Freeman, and Michael Rubinstein. Omnimatte: Associating Objects and Their Effects in Video. CVPR, 2021.