Meta-pre-training with augmentations to generalize neural network processing for domain adaptation
A method receives first and second data generated from a first and second domains including first and second set of objects, receiving first class labels for each of the first set of objects, and receiving second class labels for each of the second set of objects. The method generates a training dataset by augmenting the first data and corresponding first class labels, and locally updating neural network parameters of a model based on the training dataset. The method generates a validation dataset by augmenting the second data and corresponding second class labels, and globally updating the neural network parameters of the model based on the validation dataset. The method also generates multiple target labels for target data generated from a target domain including a third set of objects after globally updating the neural network parameters of the model based on the validation dataset.
1 . A computer-implemented method comprising:
receiving first data generated from a first domain including a first plurality of objects;
receiving first class labels for each of the first plurality of objects;
receiving second data generated from a second domain including a second plurality of objects;
receiving second class labels for each of the second plurality of objects;
generating a training dataset for each task of a set of tasks by augmenting samples of each task, the first data and corresponding first class labels comprising a first set of bounding boxes, each task representing a different augmentation strategy;
locally updating neural network parameters of a model based on the training dataset;
generating a validation dataset by augmenting the second data and corresponding second class labels comprising a second set of bounding boxes;
globally updating the neural network parameters of the model based on the validation dataset; and
generating a plurality of target labels for target data generated from a target domain including a third plurality of objects, the target labels generated by the model after globally updating the neural network parameters of the model based on the validation dataset.
2 . The method of claim 1 , in which the model comprises a three-dimensional (3D) object detection model, the plurality of target labels comprise a plurality of target bounding boxes that localize each of the third plurality of objects defined by the target data.
3 . The method of claim 2 , in which augmenting the first data and the corresponding first class labels comprises scaling the first set of bounding boxes, and augmenting the second data and the corresponding second class labels comprises scaling the second set of bounding boxes.
4 . The method of claim 1 , in which a first distribution for augmenting the first data and the corresponding first class labels differs from a second distribution for augmenting the second data and the corresponding second class labels.
5 . The method of claim 1 , in which the locally updating occurs based on each class type of the first class labels within an inner loop and the globally updating occurs based on each class type of the second class labels within an outer loop.
6 . The method of claim 1 , in which the first domain comprises a first environment, the first data comprises first point cloud data, the second domain comprises a second environment, the second data comprises second point cloud data, the target domain comprises a target environment, and the target data comprises target point cloud data.
7 . An apparatus, comprising:
at least one memory; and
at least one processor coupled to the at least one memory, the at least one processor configured:
to receive first data generated from a first domain including a first plurality of objects;
to receive first class labels for each of the first plurality of objects;
to receive second data generated from a second domain including a second plurality of objects;
to receive second class labels for each of the second plurality of objects;
to generate a training dataset for each task of a set of tasks by augmenting samples of each task, the first data and corresponding first class labels comprising a first set of bounding boxes, each task representing a different augmentation strategy;
to locally update neural network parameters of a model based on the training dataset;
to generate a validation dataset by augmenting the second data and corresponding second class labels comprising a second set of bounding boxes;
to globally update the neural network parameters of the model based on the validation dataset; and
to generate a plurality of target labels for target data generated from a target domain including a third plurality of objects, the target labels generated by the model after globally updating the neural network parameters of the model based on the validation dataset.
8 . The apparatus of claim 7 , in which the model comprises a three-dimensional (3D) object detection model, the plurality of target labels comprise a plurality of target bounding boxes that localize each of the third plurality of objects defined by the target data.
9 . The apparatus of claim 8 , in which the at least one processor is further configured to augment the first data and the corresponding first class labels by scaling the first set of bounding boxes, and further configured to augment the second data and the corresponding second class labels by scaling the second set of bounding boxes.
10 . The apparatus of claim 7 , in which a first distribution for augmenting the first data and the corresponding first class labels differs from a second distribution for augmenting the second data and the corresponding second class labels.
11 . The apparatus of claim 7 , in which the at least one processor is further configured to locally update the neural network parameters based on each class type of the first class labels within an inner loop and configured to globally update based on each class type of the second class labels within an outer loop.
12 . The apparatus of claim 7 , in which the first domain comprises a first environment, the first data comprises first point cloud data, the second domain comprises a second environment, the second data comprises second point cloud data, the target domain comprises a target environment, and the target data comprises target point cloud data.
13 . A non-transitory computer-readable medium having program code recorded thereon, the program code executed by a processor and comprising:
program code to receive first data generated from a first domain including a first plurality of objects;
program code to receive first class labels for each of the first plurality of objects;
program code to receive second data generated from a second domain including a second plurality of objects;
program code to receive second class labels for each of the second plurality of objects;
program code to generate a training dataset for each task of a set of tasks by augmenting samples of each task, the first data and corresponding first class labels comprising a first set of bounding boxes, each task representing a different augmentation strategy;
program code to locally update neural network parameters of a model based on the training dataset;
program code to generate a validation dataset by augmenting the second data and corresponding second class labels comprising a second set of bounding boxes;
program code to globally update the neural network parameters of the model based on the validation dataset; and
program code to generate a plurality of target labels for target data generated from a target domain including a third plurality of objects, the target labels generated by the model after globally updating the neural network parameters of the model based on the validation dataset.
14 . The non-transitory computer-readable medium of claim 13 , in which the model comprises a three-dimensional (3D) object detection model, the plurality of target labels comprise a plurality of target bounding boxes that localize each of the third plurality of objects defined by the target data.
15 . The non-transitory computer-readable medium of claim 14 , in which the program code to generate the training dataset by augmenting the first data and the corresponding first class labels further comprises program code to scale the first set of bounding boxes, and the program code to generate the validation dataset by augmenting the second data and corresponding second class labels further comprises program code to scale the second set of bounding boxes.
16 . The non-transitory computer-readable medium of claim 13 , in which a first distribution for augmenting the first data and the corresponding first class labels differs from a second distribution for augmenting the second data and the corresponding second class labels.
17 . The non-transitory computer-readable medium of claim 13 , in which the program code to locally update is based on each class type of the first class labels within an inner loop and the program code to globally update updates based on each class type of the second class labels within an outer loop.
18 . The non-transitory computer-readable medium of claim 13 , in which the first domain comprises a first environment, the first data comprises first point cloud data, the second domain comprises a second environment, the second data comprises second point cloud data, the target domain comprises a target environment, and the target data comprises target point cloud data.
19 . An apparatus for a computer-implemented method, comprising:
means for receiving first data generated from a first domain including a first plurality of objects;
means for receiving first class labels for each of the first plurality of objects;
means for receiving second data generated from a second domain including a second plurality of objects;
means for receiving second class labels for each of the second plurality of objects;
means for generating a training dataset for each task of a set of tasks by augmenting samples of each task, the first data and corresponding first class labels comprising a first set of bounding boxes, each task representing a different augmentation strategy;
means for locally updating neural network parameters of a model based on the training dataset;
means for generating a validation dataset by augmenting the second data and corresponding second class labels comprising a second set of bounding boxes;
means for globally updating the neural network parameters of the model based on the validation dataset; and
means for generating a plurality of target labels for target data generated from a target domain including a third plurality of objects, the target labels generated by the model after globally updating the neural network parameters of the model based on the validation dataset.
20 . The apparatus of claim 19 , in which the model comprises a three-dimensional (3D) object detection model, the plurality of target labels comprise a plurality of target bounding boxes that localize each of the third plurality of objects defined by the target data.
21 . The apparatus of claim 20 , in which the means for generating the training dataset by augmenting the first data and the corresponding first class labels comprises means for scaling the first set of bounding boxes, and the means for generating the validation dataset by augmenting the second data and corresponding second class labels comprises means for scaling the second set of bounding boxes.
22 . The apparatus of claim 19 , in which a first distribution for augmenting the first data and the corresponding first class labels differs from a second distribution for augmenting the second data and the corresponding second class labels.
23 . The apparatus of claim 19 , in which the means for locally updating is based on each class type of the first class labels within an inner loop and the means for globally updating is based on each class type of the second class labels within an outer loop.
24 . The apparatus of claim 19 , in which the first domain comprises a first environment, the first data comprises first point cloud data, the second domain comprises a second environment, the second data comprises second point cloud data, the target domain comprises a target environment, and the target data comprises target point cloud data.