This is an initial implementation of Multi-stage Deep Classifier Cascades (MDCC) for Open World Recognition task as described in our paper:
Multi-stage Deep Classifier Cascades for Open World Recognition. The 28th ACM International Conference on Information and Knowledge Management (CIKM 2019) Xiaojie Guo, Amir Alipour-Fanid, Lingfei Wu, Hemant Purohit, Xiang Chen, Kai Zeng and Liang Zhao.
Datasets from this paper
The datasets are generated by the coauthor and do please cite formally when you use these two dataset as:
Guo X, Alipour-Fanid A, Wu L, Purohit H, Chen X, Zeng K, Zhao L. Multi-stage Deep Classifier Cascades for Open World Recognition. InProceedings of the 28th ACM International Conference on Information and Knowledge Management 2019 Nov 3 (pp. 179-188).
Hemant Purohit, Carlos Castillo, Fernando Diaz, Amit Sheth, and Patrick Meier. 2014. Emergency-relief coordination on social media: Automatically matching resource requests and offers. First Monday 19, 1 (2014)
Additional dataset in open world domain
ILSVRC2010. It contains about 1.2M images for training (with 650 ∼ 3000 images per class), 50K images for validation and 150K images for testing.
Places-2. The second dataset is the recent Places-2 dataset, which contains over 10M images of 400 different scene types. The dataset features 5000 to 30,000 training images per class, consistent with real-world frequencies of occurrence.
Run the model
For RF task, when run the model:
For twotter task, run the model as:
The training procedure, testing procedure, and ealuating procedure are self-contained in the
You are welcom to connect me regarding any confusions about the codes.