Check out our latest research on weakly-supervised 3D shape completion.


Jia-Xin Zhao, Ren Bo, Qibin Hou, Ming-Ming Cheng. FLIC: Fast Linear Iterative Clustering with Active Search. CoRR, 2016.

Zhao et al. introduce a variant of SLIC superpixels [2] called FLIC – Fast Linear Iterative Clustering with Active Search. Instead of having a local search area around each superpixel, where we need to compute distance for all pixels, they let each pixel individually decide on possible labels based on the neighboring pixels. The actual algorithm is very similar to SEEDS [2] (on the last level). Essentially, all pixels are traversed (in a specific order to improve efficiency, see paper) and for each neighboring pixel with a different label, the pixel computes the distance to this superpixel and might change its label. If the pixel changes label, the corresponding superpixels (I.e the centroids) are updated on the fly – this avoids alternating between assignment and update steps as done in SLIC. However, this scheme is also employed in SEEDS.

Unfortunately, the paper only provides few qualitative examples in comparison to other approaches; therefore, Figure 1 only shows results of the proposed method. For quantitative results, see the paper. Overall, I find that the results look very similar to SLIC, and the approach is very similar to SEEDS …

Figure 1: Qualitative results.

  • [1] Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurélien Lucchi, Pascal Fua, Sabine Süsstrunk. SLIC Superpixels Compared to State-of-the-Art Superpixel Methods. IEEE Trans. Pattern Anal. Mach. Intell. 34(11): 2274-2282 (2012).
  • [2] Michael Van den Bergh, Xavier Boix, Gemma Roig, Benjamin de Capitani, Luc J. Van Gool. SEEDS: Superpixels Extracted via Energy-Driven Sampling. ECCV (7) 2012: 13-26.

What is your opinion on the summarized work? Or do you know related work that is of interest? Let me know your thoughts in the comments below or get in touch with me: