Closing the Loop for Edge Detection and Object Proposals
Authors: Yao Lu, Linda Shapiro
AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In the experiments we show that by closing the loop for edge detection and object proposals, we observe improvements for both tasks. Quantitative Evaluations To follow the object proposal evaluation routine, we leverage the Pascal VOC 2012 (Everingham et al. 2010) dataset with RGB images, and the NYU-D (Silberman et al. 2012) dataset with RGB-D images. |
| Researcher Affiliation | Academia | Yao Lu, Linda Shapiro University of Washington {luyao, shapiro}@cs.washington.edu |
| Pseudocode | Yes | Algorithm 1: Grouping proposal boxes Name: cluster CCs(CP ) Steps: h , C ; sort CP by area in descending order. for pcc in CP do if pcc / C and c C, pcc / c.children, then C = C pcc. for c C do if |c.children| > 1, then h h cluster CCs(CP C). else h = h {c}. return {h} in a tree structure. |
| Open Source Code | No | No explicit statement or link providing access to the source code for the described methodology was found. |
| Open Datasets | Yes | To follow the object proposal evaluation routine, we leverage the Pascal VOC 2012 (Everingham et al. 2010) dataset with RGB images, and the NYU-D (Silberman et al. 2012) dataset with RGB-D images. |
| Dataset Splits | Yes | To follow the object proposal evaluation routine, we leverage the Pascal VOC 2012 (Everingham et al. 2010) dataset with RGB images, and the NYU-D (Silberman et al. 2012) dataset with RGB-D images. For the later dataset we apply the training-testing split from (Gupta et al. 2014), which contains 795 training images. Quantitative evaluation of object proposals on the Pascal VOC2012 validation (RGB) and the NYU-D test (RGB-D) datasets. |
| Hardware Specification | No | No specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments were provided. |
| Software Dependencies | No | The paper mentions using specific external tools like the 'Structured Edge detector (Doll ar and Zitnick 2013)' and 'Edge Box (Zitnick and Doll ar 2014)', but does not provide version numbers for these or other software dependencies. |
| Experiment Setup | Yes | We set the search radius r > 1, so that disconnected edges can be found during the search. To initialize the algorithm, we randomly pick a starting terminal point near the bounding box boundary. Meanwhile we set λ = 0.8 to put more weight on the size factor. We set a threshold on the objectness score before grouping the closed contours. This is to reduce noise from pool-ranked candidate boxes. |