Nonparametric Scene Parsing with Adaptive Feature Relevance and Semantic Context
read more
Citations
Towards unified depth and semantic prediction from a single image
Learning to segment under various forms of weak supervision
ReSeg: A Recurrent Neural Network-Based Model for Semantic Segmentation
Context Driven Scene Parsing with Attention to Rare Classes
DAG-Recurrent Neural Networks for Scene Labeling
References
Distinctive Image Features from Scale-Invariant Keypoints
Histograms of oriented gradients for human detection
Distinctive Image Features from Scale-Invariant Keypoints
Beyond Bags of Features: Spatial Pyramid Matching for Recognizing Natural Scene Categories
Modeling the Shape of the Scene: A Holistic Representation of the Spatial Envelope
Related Papers (5)
Frequently Asked Questions (14)
Q2. What are the contributions in "Nonparametric scene parsing with adaptive feature relevance and semantic context" ?
This paper presents a nonparametric approach to semantic parsing using small patches and simple gradient, color and location features. To further improve the accuracy of the nonparametric approach, the authors examine the importance of the retrieval set used to compute the nearest neighbours using a novel semantic descriptor to retrieve better candidates.
Q3. What future works have the authors mentioned in the paper "Nonparametric scene parsing with adaptive feature relevance and semantic context" ?
For future work, the authors would like to explore better methods for incorporating spatial information at the patch level and also explore learning semantic concepts for scene understanding.
Q4. What is the problem of semantic labelling?
The problem of semantic labelling, requires simultaneous segmentation of an image into regions and categorization of all the image pixels.
Q5. What is the metric used to compute the relevance of a feature channel i?
For the query point x0, the relevance for feature i can be computed by averaging the ri(z)’s in its neighbourhoodr̄i(x0) = 1 |N(x0)| ∑z∈N(x0) ri(z) (10)where N(x0) denotes a neighbourhood centered at x0 (using the current feature weights) with K0 points in it.
Q6. How many iterations of the weight computation step in Eq. (11)?
The authors carry out 5 iterations of the weight computation step in Eq. (11) adaptively changing the nearest neighbours in the weighted neighbourhood space.
Q7. What are the categories which saw an increase of more than 10% after the use of semantic context?
The categories which saw an increase of more than 10% after the use of semantic context include field, car, river, plant, sidewalk, bridge, door, crosswalk.
Q8. How do the authors compute the appearance likelihood for the entire image?
In order to compute the appearance likelihood for the entire image, the authors approximate the Naive Bayes assumption yieldingP (A|L) ≈ S∏i=1P (ai|li).
Q9. What other datasets were used to evaluate the performance of their method?
For evaluating the performance of their method, the authors tested and compared it with several state-of-the-art techniques on four different datasets: SiftFlow [15], SUN09 [1], Google Street View [30] and Stanford Background [6].
Q10. What is the metric used to compute the weight of a query image?
With the varying nature of the retrieval set for individual query images, the authors use the locally adaptive metric approach of [3] for the weight computation.
Q11. What is the semantic label descriptor for a labelled image?
Their proposed descriptor helps encode the positional information of each category in the image and can be used for semantic contextual retrieval.
Q12. How do the authors perform semantic labelling on the image?
Using the new retrieval set Ts, the authors once again perform semantic labelling on the image by the process described in Section 3.3- 3.4.
Q13. How was the work of [26] extended?
The work of [26] was extended by [4] by training per superpixel per feature weights and also by incorporating superpixellevel semantic context.
Q14. What is the posterior probability of a labelling L given the observed appearance vectors?
The posterior probability of a labelling L given the observed appearance feature vectors A = [a1, a2, . . . ,aS ] computed for each superpixel can be expressed as:P (L|A) = P (A|L)P (L) P (A) .