Graph-based color description generation

1. A computer-implemented method for generating a description in natural language for a color, the computer-implemented method comprising:

acquiring a list of tuples, each of the tuples including a color and a description;

generating a graph by using each of the tuples as a node and adding edges between the nodes when a difference between colors of the nodes in terms of human perception is outside a predetermined range;

filtering the edges based on external color comparative descriptions stored in an external color comparative database;

incorporating a new node in the graph by finding a closest neighbor node based on the color difference in terms of human perception and adding a new edge between the new node and the closest neighbor node, the new node including a new color which is not included in the list of tuples and has no description;

learning a feature vector for each of the nodes in the graph by using message passing and the colors of the nodes as initial seeds; and

generating a description of the new node by using each of the learned feature vectors as initial states for a recurrent neural network (RNN)-based decoder in a semi-supervised fashion.

2. The computer-implemented method of claim 1 , wherein filtering the edges involves removing redundant or unfeasible edges from the graph.

3. The computer-implemented method of claim 1 , wherein the color difference in terms of human perception is provided by metric Delta E.

4. The computer-implemented method of claim 3 , wherein, if the Delta E between two colors is between 11 and 49, an edge is generated between the two colors.

5. The computer-implemented method of claim 1 , wherein, after the new node is added to the graph, contextual information from the edges is used from a set of labeled nodes to a set of unlabeled nodes to learn a description for the new node.

6. The computer-implemented method of claim 1 , wherein loss in the RNN is computed on a labeled portion of the graph.

7. The computer-implemented method of claim 1 , wherein the new node is defined as an unlabeled node.

8. A computer-implemented method for generating a description in natural language for a color, the computer-implemented method comprising:

obtaining a list of tuples, each of the tuples including a color and a description;

generating a graph by using each of the tuples as a node and adding edges between the nodes based on a similarity threshold;

filtering the edges based on external color comparative descriptions stored in an external color comparative database; and

incorporating a new node in the graph by finding a closest neighbor node based on the similarity threshold and adding a new edge between the new node and the closest neighbor node.

9. The computer-implemented method of claim 8 , wherein the new node includes a new color not included in the list of tuples and has no description.

10. The computer-implemented method of claim 9 , further comprising learning a feature vector for each of the nodes in the graph.

11. The computer-implemented method of claim 10 , wherein the feature vector is learned by using message passing and the colors of the nodes as initial seeds.

12. The computer-implemented method of claim 11 , further comprising generating a description of the new node by designating each of the learned feature vectors as initial states.

13. The computer-implemented method of claim 12 , wherein the initial states are fed into a recurrent neural network (RNN)-based decoder in a semi-supervised fashion.

14. The computer-implemented method of claim 13 , wherein the similarity threshold is a difference between colors of the nodes in terms of human perception outside a predetermined range.

15. The computer-implemented method of claim 14 , wherein the color difference in terms of human perception is provided by metric Delta E.

16. The computer-implemented method of claim 15 , wherein, if the Delta E between two colors is between 11 and 49, an edge is generated between the two colors.

17. The computer-implemented method of claim 16 , wherein loss in the RNN is computed on a labeled portion of the graph.

18. The computer-implemented method of claim 8 , wherein filtering the edges involves removing redundant or unfeasible edges from the graph.

19. The computer-implemented method of claim 8 , wherein, after the new node is added to the graph, contextual information from the edges is used from a set of labeled nodes to a set of unlabeled nodes to learn a description for the new node.

20. A computer program product for generating a description in natural language for a color, the computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to:

acquire a list of tuples, each of the tuples including a color and a description;

generate a graph by using each of the tuples as a node and add edges between the nodes when a difference between colors of the nodes in terms of human perception is outside a predetermined range;

filter the edges based on external color comparative descriptions stored in an external color comparative database;

incorporate a new node in the graph by finding a closest neighbor node based on the color difference in terms of human perception and add a new edge between the new node and the closest neighbor node, the new node including a new color which is not included in the list of tuples and has no description;

learn a feature vector for each of the nodes in the graph by using message passing and the colors of the nodes as initial seeds; and

generate a description of the new node by using each of the learned feature vectors as initial states for a recurrent neural network (RNN)-based decoder in a semi-supervised fashion.

21. The computer program product of claim 20 , wherein, after the new node is added to the graph, contextual information from the edges is used from a set of labeled nodes to a set of unlabeled nodes to learn a description for the new node.

22. A computer program product for generating a description in natural language for a color, the computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to:

obtain a list of tuples, each of the tuples including a color and a description;

generate a graph by using each of the tuples as a node and add edges between the nodes based on a similarity threshold;

filter the edges based on external color comparative descriptions stored in an external color comparative database; and

incorporate a new node in the graph by finding a closest neighbor node based on the similarity threshold and add a new edge between the new node and the closest neighbor node.

23. The computer program product of claim 22 , wherein a feature vector is learned for each of the nodes in the graph, the feature vector learned by using message passing and the colors of the nodes as initial seeds.

24. The computer program product of claim 23 , wherein a description of the new node is generated by designating each of the learned feature vectors as initial states fed into a recurrent neural network (RNN)-based decoder in a semi-supervised fashion.

25. A computer-implemented method for generating a description in natural language for a color, the computer-implemented method comprising:

generating a graph to structure color relationships, the graph including a plurality of nodes and edges;

removing redundant edges based on color comparative descriptions;

incorporating new nodes with new edges in the graph based on a similarity threshold;

learning a feature vector for each of the nodes of the graph; and

generating a description of the new nodes by employing the feature vectors, the feature vectors fed into a neural network in a semi-supervised manner.