Methods and systems for training an object detection algorithm using synthetic images
A non-transitory computer readable medium embodies instructions that cause one or more processors to perform a method. The method includes selecting a 3D model corresponding to an object. The method further includes generating domain-adapted images of the 3D model, the domain-adapted images representing the 3D model at corresponding poses. The method further includes acquiring 2D projections of 3D points on a 3D bounding box defined around the 3D model at the corresponding poses. The method further includes training an algorithm model to learn correspondences between the generated images and the corresponding 2D projections. The method further includes storing, in a memory, parameters representing the algorithm model.
1. A non-transitory computer readable medium that embodies instructions that cause one or more processors to perform a method comprising:
(a) selecting a 3D model corresponding to an object;
(b) generating domain-adapted images of the 3D model, the domain-adapted images representing the 3D model at corresponding poses, including:
(b1) providing the 3D model with information representing randomly or algorithmically chosen or generated textures that are different from one another; and
(b2) rendering the 3D model at the corresponding poses to obtain the domain-adapted images, at least two of the domain-adapted images representing the 3D model with the textures that are different between the at least two of the domain-adapted images;
(c) acquiring 2D projections of 3D points on a 3D bounding box defined around the 3D model at the corresponding poses;
(d) training an algorithm model to learn correspondences between the generated images and the corresponding 2D projections; and
(e) storing, in a memory, parameters representing the algorithm model.
2. The non-transitory computer readable medium according to claim 1 , wherein the randomly or algorithmically chosen or generated texture is applied to a background or the rendering of the 3D model.
3. The non-transitory computer readable medium according to claim 1 , wherein classification information for each of the domain-adapted images is included in the algorithm model.
4. The non-transitory computer readable medium according to claim 1 , wherein (b) generating domain-adapted images includes:
(b3) rendering the 3D model at the corresponding poses to obtain pre-images; and
(b4) applying an enhancement filter to the pre-images to obtain the domain-adapted images.
5. A non-transitory computer readable medium that embodies instructions that cause one or more processors to perform a method comprising:
(a) selecting a 3D model corresponding to an object;
(b) generating domain-adapted images of the 3D model, the images representing the 3D model at corresponding poses, including:
(b1) providing the 3D model with information representing randomly or algorithmically chosen or generated textures that are different from one another; and
(b2) rendering the 3D model at the corresponding poses to obtain the domain-adapted images, at least two of the domain-adapted images representing the 3D model with the textures that are different between the at least two of the domain-adapted images;
(c) training an algorithm model to learn correspondences between the acquired images and corresponding classes, the classes representing respective orientations of the 3D model; and
(d) storing, in a memory, parameters representing the algorithm model.
6. The non-transitory computer readable medium according to claim 5 , wherein the randomly or algorithmically chosen or generated texture is applied to a background or the rendering of the 3D model.
7. The non-transitory computer readable medium according to claim 5 , wherein (b) generating domain-adapted images includes:
(b3) rendering the 3D model at the corresponding poses to obtain pre-images; and
(b4) applying an enhancement filter to the pre-images to obtain the domain-adapted images.
8. A method comprising:
(a) using one or more processors, selecting a 3D model corresponding to an object;
(b) using the one or more processors, generating domain-adapted images of the 3D model, the domain-adapted images representing the 3D model at corresponding poses, including:
(b1) providing the 3D model with information representing randomly or algorithmically chosen or generated textures that are different from one another; and
(b2) rendering the 3D model at the corresponding poses to obtain the domain-adapted images, at least two of the domain-adapted images representing the 3D model with the textures that are different between the at least two of the domain-adapted images;
(c) using the one or more processors, acquiring 2D projections of 3D points on a 3D bounding box defined around the 3D model at the corresponding poses;
(d) using the one or more processors, training an algorithm model to learn correspondences between the generated images and the corresponding 2D projections; and
(e) storing, in a memory, parameters representing the algorithm model.
9. The method according to claim 8 , wherein the randomly or algorithmically chosen or generated texture is applied to a background or the rendering of the 3D model.
10. The method according to claim 8 , wherein classification information for each of the domain-adapted images is included in the algorithm model.
11. The method according to claim 8 , wherein (b) generating domain-adapted images includes:
(b3) rendering the 3D model at the corresponding poses to obtain pre-images; and
(b4) applying an enhancement filter to the pre-images to obtain the domain-adapted images.
12. A method comprising:
(a) using one or more processors, selecting a 3D model corresponding to an object;
(b) using the one or more processors, generating domain-adapted images of the 3D model, the images representing the 3D model at corresponding poses, including:
(b1) providing the 3D model with information representing randomly or algorithmically chosen or generated textures that are different from one another; and
(b2) rendering the 3D model at the corresponding poses to obtain the domain-adapted images, at least two of the domain-adapted images representing the 3D model with the textures that are different between the at least two of the domain-adapted images;
(c) using the one or more processors, training an algorithm model to learn correspondences between the acquired images and corresponding classes, the classes representing respective orientations of the 3D model; and
(d) storing, in a memory, parameters representing the algorithm model.
13. The method according to claim 12 , wherein the randomly or algorithmically chosen or generated texture is applied to a background or the rendering of the 3D model.
14. The method according to claim 12 , wherein (b) generating domain-adapted images includes:
(b3) rendering the 3D model at the corresponding poses to obtain pre-images; and
(b4) applying an enhancement filter to the pre-images to obtain the domain-adapted images.