Supplementary Material

Seeing 3D chairs: exemplar part-based 2D-3D alignment using a large dataset of CAD models

Mathieu Aubry, Daniel Maturana, Alexei A. Efros, Bryan Russell, Josef Sivic

1. User study results

Here we show the complete set of results evaluated in the user-study described in the paper (table 1). The evaluated output for our algorithm can be seen here and those for the exemplar-LDA baseline can be accessed here. Note that for our algorithm we show results up to 50% recall, whereas the user-study was performed only on results up to 25% recall due to the limited recall of the baseline.

Similar results on the dataset of 229 images containing 374 chairs from the Pascal VOC 2007 test set (not shown in the paper) can be accessed here.

2. False positives

To illustrate the different types of errors, the top 25 false positives on the Pascal VOC 2012 chair dataset can be seen here.

3. Top alignments for detected chairs

Similar to figure 4 in the paper, we show here the top 10 aligned chairs returned by our algorithm for the top 20 detected images in the Pascal VOC 2012 chairs dataset.