Active Detection via Adaptive Submodularity
Authors: Yuxin Chen, Hiroaki Shioi, Cesar Fuentes Montesinos, Lian Pin Koh, Serge Wich, Andreas Krause
ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the proposed algorithm on three real-world tasks, including a problem for biodiversity monitoring from micro UAVs in the Sumatra rain forest. Our results show that active detection not only outperforms its passive counterpart; for certain tasks, it also works significantly better than straightforward application of existing active learning techniques. |
| Researcher Affiliation | Academia | ETH Z urich, Z urich, Switzerland The University of Tokyo, Tokyo, Japan * Liverpool John Moores University, Liverpool, United Kingdom |
| Pseudocode | Yes | Algorithm 1 The active detection algorithm |
| Open Source Code | No | The paper does not contain an explicit statement about the release of source code for the described methodology, nor does it provide any links to a code repository. |
| Open Datasets | Yes | To estimate the distribution of critically endangered Sumatran orangutans (Pongo abelii), ecologists deploy conservation drones above orangutan habitat in surveyed areas, so that they can obtain timely and high-quality photographs of orangutan nests high in the tree canopies (Koh & Wich, 2012). ... We apply Alg. 1 to the TUD-crossing sequence, based on the Hough Forest detector proposed in Gall & Lempitsky (2009). ... Object detection on PASCAL VOC Data Set |
| Dataset Splits | Yes | As we do not have sufficient (positive) training data, we use all the labeled images other than those in the current test image as training set. ... We test the candidate algorithms on 41 frames of the TUD-crossing sequence (by sampling every 5th frame of the full video sequence) |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU models, CPU types, memory) used to conduct the experiments. |
| Software Dependencies | No | The paper mentions various algorithms and models (e.g., linear discriminant classifier (LDA), k-means, Hough Forest detector, Hungarian algorithm, deformable parts model (MDPM)), but it does not specify any version numbers for these or other software dependencies. |
| Experiment Setup | Yes | Each training example is represented as a 9-d vector which consists of statistics (mean, maximum and minimum) of three color channels in a patch. Based on these features, we train a linear discriminant classifier (LDA) in order to classify orangutan nests vs. background. ... We use a discount factor γ = 0.01 to penalize votes that are similar with any of the incorrect votes. ... We limited the maximum number of detections to be 10 for both systems (given there are at most 8 objects per frame) in order to have a fair comparison. ... Each detector makes 16 detections per image. |