Semi-supervised person re-identification using multi-view clustering
A semi-supervised model incorporates deep feature learning and pseudo label estimation into a unified framework. The deep feature learning can include multiple convolutional neural networks (CNNs). The CNNs can be trained on available training datasets, tuned using a small amount of labeled training samples, and stored as the original models. Features are then extracted for unlabeled training samples by utilizing the original models. Multi-view clustering is used to cluster features to generate pseudo labels. Then the original models are tuned by using an updated training set that includes labeled training samples and unlabeled training samples with pseudo labels. Iterations of multi-view clustering and tuning using an updated training set can continue until the updated training set is stable.
1. A computer system, comprising:
a processor; and
system memory coupled to the processor and storing instructions configured to cause the processor to:
initialize a set of person re-identification neural networks on labeled training data;
iteratively tune the set of person re-identification neural networks using unlabeled data along with pseudo labels estimated for the unlabeled data; and
after tuning:
extract human facial features from the unlabeled data set;
perform multi-view clustering on the extracted human facial features estimating additional pseudo labels for the unlabeled data set including calculating an identification loss and a ranking loss per extracted human facial feature; and
further tune the set of person re-identification neural networks using the unlabeled data along with the additional pseudo labels.
2. The computer system of claim 1 , wherein the set of neural networks comprises a set of convolutional neural networks.
3. The computer system of claim 1 , wherein instructions configured to initialize the set of neural networks on labeled training data comprises instructions configured to initialize the set of neural networks on ImageNet data.
4. The computer system of claim 1 , further comprising instructions configured to:
extract features from the unlabeled data set; and
perform multi-view clustering on the extracted features estimating the pseudo labels.
5. The computer system of claim 4 , wherein instructions configured to perform multi-view clustering on the extracted features comprise instructions configured to calculate an identification loss and a ranking loss per extracted feature.
6. The computer system of claim 5 , wherein instructions configured to extract features from the unlabeled data set comprise instructions configured to extract additional human facial features from the unlabeled data set.
7. The computer system of claim 1 , wherein instructions configured to perform multi-view clustering on the extracted human facial features estimating additional pseudo labels for the unlabeled data set comprise instructions configured to calculate a centroid matrix for the extracted human facial features based on a weight factor and weight distribution control parameter.
8. The computer system of claim 7 , wherein instructions configured to calculate a centroid matrix for the extracted human facial features based on a weight factor and weight distribution control parameter comprise instructions configured to calculate a centroid matrix satisfying a coding scheme and being an expected number of clusters.
9. The computer system of claim 1 , wherein instructions configured to perform multi-view clustering on the extracted human facial features estimating additional pseudo labels comprise instructions configured to re-estimate the pseudo labels.
10. The computer system of claim 1 , wherein instructions configured to iteratively tune the set of person re-identification neural networks comprise instructions configured to iteratively tune the set of person re-identification neural networks with pseudo labels estimated from human facial features extracted from the unlabeled data.
11. A method, comprising:
initializing a set of person re-identification neural networks on labeled training data;
iteratively tuning the set of person re-identification neural networks using unlabeled data along with pseudo labels estimated for the unlabeled data; and
after tuning:
extracting human facial features from the unlabeled data set;
performing multi-view clustering on the extracted human facial features estimating additional pseudo labels for the unlabeled data set including calculating an identification loss and a ranking loss per extracted human facial feature; and
further tuning the set of person re-identification neural networks using the unlabeled data along with the additional pseudo labels.
12. The method of claim 11 , wherein the set of neural networks comprises a set of convolutional neural networks.
13. The method of claim 11 , wherein initializing the set of neural networks on labeled training data comprises initializing the set of neural networks on ImageNet data.
14. The method of claim 11 , further comprising:
extracting features from the unlabeled data set; and
performing multi-view clustering on the extracted features estimating the pseudo labels.
15. The method of claim 14 , wherein performing multi-view clustering on the extracted features comprises calculating an identification loss and a ranking loss per extracted feature.
16. The method of claim 15 , wherein extracting features from the unlabeled data set comprises extracting additional human facial features from the unlabeled data set; and
wherein performing multi-view clustering on the extracted features estimating the pseudo labels comprises performing multi-view clustering on the extracted human facial features.
17. The method of claim 11 , wherein performing multi-view clustering on the extracted human facial features estimating additional pseudo labels for the unlabeled data set comprises calculating a centroid matrix for the extracted human facial features based on a weight factor and weight distribution control parameter.
18. The method of claim 17 , wherein calculating a centroid matrix for the extracted human facial features based on a weight factor and weight distribution control parameter comprises calculating a centroid matrix satisfying a coding scheme and being an expected number of clusters.
19. The method of claim 11 , wherein performing multi-view clustering on the extracted human facial features estimating additional pseudo labels comprises re-estimating the pseudo labels.
20. The computer system of claim 11 , wherein iteratively tuning the set of person re-identification neural networks comprises iteratively tuning the set of person re-identification neural networks with pseudo labels estimated from human facial features extracted from the unlabeled data.