Spatial Consistency Loss for Training Multi-Label Classifiers from Single-Label Annotations
0
Citation
0
Reference
10
Related Paper
Abstract:
As natural images usually contain multiple objects, multi-label image classification is more applicable "in the wild" than single-label classification. However, exhaustively annotating images with every object of interest is costly and time-consuming. We aim to train multi-label classifiers from single-label annotations only. We show that adding a consistency loss, ensuring that the predictions of the network are consistent over consecutive training epochs, is a simple yet effective method to train multi-label classifiers in a weakly supervised setting. We further extend this approach spatially, by ensuring consistency of the spatial feature maps produced over consecutive training epochs, maintaining per-class running-average heatmaps for each training image. We show that this spatial consistency loss further improves the multi-label mAP of the classifiers. In addition, we show that this method overcomes shortcomings of the "crop" data-augmentation by recovering correct supervision signal even when most of the single ground truth object is cropped out of the input image by the data augmentation. We demonstrate gains of the consistency and spatial consistency losses over the binary cross-entropy baseline, and over competing methods, on MS-COCO and Pascal VOC. We also demonstrate improved multi-label classification mAP on ImageNet-1K using the ReaL multi-label validation set.Keywords:
Pascal (unit)
Ground truth
Multi-label classification
Training set
Contextual image classification
Feature (linguistics)
We report on the methods used in our recent DeepEnsembleCoco submission to the PASCAL VOC 2012 challenge, which achieves state-of-the-art performance on the object detection task. Our method is a variant of the R-CNN model proposed Girshick:CVPR14 with two key improvements to training and evaluation. First, our method constructs an ensemble of deep CNN models with different architectures that are complementary to each other. Second, we augment the PASCAL VOC training set with images from the Microsoft COCO dataset to significantly enlarge the amount training data. Importantly, we select a subset of the Microsoft COCO images to be consistent with the PASCAL VOC task. Results on the PASCAL VOC evaluation server show that our proposed method outperform all previous methods on the PASCAL VOC 2012 detection task at time of submission.
Pascal (unit)
Training set
Cite
Citations (46)
Training set
Multi-label classification
Cite
Citations (12)
Multi_label learning and application is a new hot issue in machine learning and data mining recently. In multi_label learning, the training set is composed of instances each associated with a set of labels, and the task is to predict the label sets of unseen instances through analyzing training instances with known label sets. In this paper, authors research on classifying multi_label data based on Naïve Bayes Classifier(NBC), which is extended to multi_label learning. Training and testing procedures are adapted to the characteristics and assessment criteria of multi_label learning problem. The adapted NBC is realized through programming on MBNC experimental platform and applied to the nature scene classification, the results show that it is effective.
Multi-label classification
Training set
Cite
Citations (4)
In the field of Remote Sensing Scene Classification (RSSC), multi-label classification has become necessary. However, the creation of a multi-label dataset is a laborious process due to the higher annotation costs compared to multi-class classification. In this study, we conducted a pioneering experiment in the context of partial-label classification on remote sensing datasets and aim to discuss the differences and limitations. In partial-label classification, each image is assigned some "positive" labels, which means it is annotated, and other "unknown" labels which are not determined as positive or negative. Consequently, the model is trained with limited information. We evaluated the classification performance on MLRSNet and AID multilabel datasets, using the method and loss functions that have shown excellent performance in previous studies on ground-level view datasets. Our code is available at https://github.com/Kf-7070/ IGARSS2023_partial_label.
Contextual image classification
Multi-label classification
Code (set theory)
Cite
Citations (0)
The machine learning has many capabilities one of them is classification. Classification employed in many contexts like telling hotel reviews good / bad, or inferring the image consists of dog, cat etc. As the data increases there is a need to organize it, for that purpose classification can be helpful. Modern classification problems involve the prediction of multiple labels simultaneously associated with a single instance known as "Multi Label Classification". In multi-label classification, each of the input data samples belongs to one or more than one classes or labels. The traditional binary and multi-class classification problems are the subset of the multi-label classification problem. In this paper we are implementing the multi label classification using CNN framework with keras libraries. Classification can be applied to different domain such as text, audio etc. In this paper we are applying classification for an image dataset.
Multi-label classification
Contextual image classification
One-class classification
Binary classification
Multiclass classification
Statistical classification
Cite
Citations (2)
In this work, the case of semantic segmentation on a small image dataset (simulated by 1000 randomly selected images from PASCAL VOC 2012), where only weak supervision signals (scribbles from user interaction) are available is studied. Especially, to tackle the problem of limited data annotations in image segmentation, transferring different pre-trained models and CRF based methods are applied to enhance the segmentation performance. To this end, RotNet, DeeperCluster, and SemiW moreover, for the case of training on the full PASCAL VOC 2012 training data, this pre-training approach increases the mIoU results by almost 4%. On the other hand, dense CRF is shown to be very effective as well, enhancing the results both as a loss regularization technique in weakly supervised training and as a post-processing tool.
Pascal (unit)
Training set
Regularization
Labeled data
Cite
Citations (0)
In this work, the case of semantic segmentation on a small image dataset (simulated by 1000 randomly selected images from PASCAL VOC 2012), where only weak supervision signals (scribbles from user interaction) are available is studied. Especially, to tackle the problem of limited data annotations in image segmentation, transferring different pre-trained models and CRF based methods are applied to enhance the segmentation performance. To this end, RotNet, DeeperCluster, and Semi&Weakly Supervised Learning (SWSL) pre-trained models are transferred and finetuned in a DeepLab-v2 baseline, and dense CRF is applied both as a post-processing and loss regularization technique. The results of my study show that, on this small dataset, using a pre-trained ResNet50 SWSL model gives results that are 7.4% better than applying an ImageNet pre-trained model; moreover, for the case of training on the full PASCAL VOC 2012 training data, this pre-training approach increases the mIoU results by almost 4%. On the other hand, dense CRF is shown to be very effective as well, enhancing the results both as a loss regularization technique in weakly supervised training and as a post-processing tool.
Pascal (unit)
Regularization
Training set
Labeled data
Cite
Citations (0)
In this paper, a novel multi-label classification model using convolutional neural networks (CNNs) is proposed. As one of the deep learning architectures, CNNs lead breakthrough in many fields of image processing especially the image classification. Since the applications of CNNs are more concentrating in the background of single-label samples, our model introduce the hidden semantic between different labels of the same sample to the existed CNNs to enhance the performance. In order to use the semantic of the multi-label, i.e. fine-label and coarse-label, a coarse-label classification part was built using the shared low features of the fine-label classification. We have compared our method with the CNNs using single label. Experimental results demonstrate that our model can achieve better classification performance on the multi-label dataset of CIFAR-100 than the CNNs using single label, our model improves the classification performance, by 2.3% increasing for the top-1 accuracy, while 2.7% for the top-5 on average.
Multi-label classification
Contextual image classification
Cite
Citations (11)
SIGPLAN and the PASCAL Users Group will jointly sponsor a session to bring all conference attendees who are interested in PASCAL together to interact. This is not a technical session in the usual sense. However, in order to convey the most information, it will consist, at least in part, of a series of short presentations on PASCAL related topics. Presentations will address varied topics related to the language including experiences with PASCAL, tools for PASCAL programming, PASCAL implementations, etc.
Pascal (unit)
Implementation
Cite
Citations (0)
ALBE/P is a language-based CRT editor for PASCAL programs. The CRT screen serves as a window through which a programmer can view and modify a “pretty-printed” picture of any part of a PASCAL program. The ALBE/P system differs from conventional screen-oriented text editors in that the program is stored as a PASCAL parse tree and the editing operations are designed specifically for the PASCAL language. Moreover, because ALBE/P is language-based, it will not accept programs with local syntax errors (e.g., ill-formed expressions) or global errors (e.g., undeclared variables).The system is also an effective tool for developing and maintaining application systems and subroutine packages in multiple language environments. Programs can be entered in Language-Neutral Form (LNF), a PASCAL subset with language features common to C, PL/I, and ALGOL. Then, ALBE will generate a PASCAL program to be run and debugged under the host operating system. When program development is complete, ALBE/LNF will generate equivalent programs for the desired target languages and operating system environments. Currently, ALBE/LNF supports PASCAL, C, and FORTRAN under VAX/VMS.
Pascal (unit)
Subroutine
Programmer
Fortran
Cite
Citations (4)