Object-Adaptive LSTM Network for Visual Tracking
read more
Citations
Object-adaptive LSTM network for real-time visual tracking with adversarial data augmentation
Improving Object Tracking by Added Noise and Channel Attention.
Two motion models for improving video object tracking performance
Object-Adaptive LSTM Network for Real-time Visual Tracking with Adversarial Data Augmentation
Multi-object Tracking Based on Nearest Optimal Template Library
References
ImageNet Large Scale Visual Recognition Challenge
TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems
High-Speed Tracking with Kernelized Correlation Filters
Online Object Tracking: A Benchmark
Object Tracking Benchmark
Related Papers (5)
Convolutional neural networks based scale-adaptive kernelized correlation filter for robust visual object tracking
Frequently Asked Questions (14)
Q2. How do the authors optimize the efficiency of their tracking method?
In order to optimize the efficiency of their tracking method, the authors propose to leverage the fast matching-based method to preevaluate the densely sampled proposals in the search region and the high-quality proposals are selected to feed to the subsequent classification network.
Q3. How can LSTM capture long-range temporal dependencies?
By taking advantage of its intrinsic recurrent structure, their network can capture long-range temporal dependencies and dynamically adapt to the object variations via memorizing the changes of object appearance and motion during online tracking.
Q4. How many convolutional layers are used in the proposed LSTM network?
In the proposed LSTM network, for the purpose of supplying the network with rich target appearance information, the authors adopt five convolutional layers pre-trained on the ILSVRC15 [15] dataset to extract high-level target features.
Q5. How do the authors fit in with the task?
To fit in with the tracking task, the authors use the LSTM network to evaluate the proposals and identify the optimal target state in each frame.
Q6. How is the LSTM recurrent structure learned online?
with a pre-trained convolutional feature extractor, the LSTM recurrence is learned online to sufficiently utilize sequence-specific information.
Q7. What is the popular benchmark for OA-LSTM?
OTB (Object Tracking benchmark) [6] is a popular tracking benchmark consisting of 100 fully annotated videos with substantial variations.
Q8. What is the way to get a confidence map?
Based on the similarity learning function, the authors can firstly obtain a confidence map, which corresponds to all the translated sub-regions in the search region.
Q9. What other trackers use deep networks to learn the object feature representations?
Among the competing trackers, SiamFC, GOTURN, CNT and CFNet also employ deep networks to learn the object feature representations and achieve good performance.
Q10. What is the LSTM network's ability to learn the variations of the target appearance and?
By taking advantage of its intrinsic recurrent structure, the LSTM network is able to capture the temporal dependencies of the video sequence and memorize the changes of target appearance and motion during online tracking.
Q11. Why are these convolutional layers kept fixed during the online learning stage?
These convolutional layers are kept fixed during the online learning stage since they have learned the capability of generic feature extraction.
Q12. What is the proposed method for tracking?
The proposed method is implemented in Python using TensorFlow [16] and runs at an average speed of 11.5 fps with 2.00GHz Intel Xeon E5-2660 and an NVIDIA GTX TITAN X GPU.
Q13. What is the way to track objects?
Experimental results on public tracking benchmarks have shown that their method achieves state-ofthe-art performance with a satisfactory speed, demonstrating the successful application of recurrent structure to visual object tracking.
Q14. What are the main limitations of the method?
Although this method has achieved outstanding accuracy, massive feature extractions and sophisticated online fine-tuning techniques limit its efficiency.