Method for diverse sequential point cloud forecasting
A method for sequential point cloud forecasting is described. The method includes training a vector-quantized conditional variational autoencoder (VQ-CVAE) framework to map an output to a closest vector in a discrete latent space to obtain a future latent space. The method also includes outputting, by a trained VQ-CVAE, a categorical distribution of a probability of V vectors in a discrete latent space in response to an input previously sampled latent space and past point cloud sequences. The method further includes sampling an inferred future latent space from the categorical distribution of the probability of the V vectors in the discrete latent space. The method also includes predicting a future point cloud sequence according to the inferred future latent space and the past point cloud sequences. The method further includes denoising, by a denoising diffusion probabilistic model (DDPM), the predicted future point cloud sequences according to an added noise.
1 . A method for sequential point cloud forecasting, comprising:
training a vector-quantized conditional variational autoencoder (VQ-CVAE) framework to map an output to a closest vector in a discrete latent space to obtain a future latent space;
outputting, by a trained VQ-CVAE, a categorical distribution of a probability of V vectors in the discrete latent space in response to an input previously sampled latent space and past point cloud sequences;
sampling an inferred future latent space from the categorical distribution of a probability of the V vectors in the discrete latent space;
predicting a future point cloud sequence according to the inferred future latent space and the past point cloud sequences; and
denoising the predicted, future point cloud sequences using a denoising diffusion probabilistic model (DDPM).
2 . The method of claim 1 , in which training comprises feeding a training encoder of the VQ-CVAE framework with a future point cloud, the input previously sampled latent space, and past point cloud sequences to predict a future latent space.
3 . The method of claim 1 , in which outputting comprises:
feeding an inference encoder of the trained VQ-CVAE with the previously sampled latent space, and past point cloud sequences;
inferring, by the inference encoder, a classification over quantized vectors; and
sampling the future latent space sampled from the output categorical distribution.
4 . The method of claim 1 , in which sampling comprises sampling, by a decoder, the future latent space from the output categorical distribution.
5 . The method of claim 1 , in which predicting comprises predicting, by a decoder, the future point cloud at time t in response to the sampled latent space and features of past point cloud sequences.
6 . The method of claim 1 , in which denoising comprises:
performing a partial denoising process on the predicted, future point cloud sequence to generate a denoised, future point cloud sequence; and
performing a partial diffusion process on the denoised, future point cloud sequence.
7 . The method of claim 6 , in which performing the partial denoising process comprises:
adding noise to a point cloud sequence sample including the predicted, future point cloud sequence and a previously predicted point cloud sequence; and
removing the noise from the point cloud sequence sample over a predetermined number of steps to provide the denoised, future point cloud sequence.
8 . The method of claim 1 , further comprising planning a trajectory of an ego vehicle according to a predicted, future point cloud sequence of agents within a scene surrounding the ego vehicle.
9 . A non-transitory computer-readable medium having program code recorded thereon for sequential point cloud forecasting, the program code being executed by a processor and comprising:
program code to train a vector-quantized conditional variational autoencoder (VQ-CVAE) framework to map an output to a closest vector in a discrete latent space to obtain a future latent space;
program code to output, by a trained VQ-CVAE, a categorical distribution of a probability of V vectors in the discrete latent space in response to an input previously sampled latent space and past point cloud sequences;
program code to sample an inferred future latent space from the categorical distribution of a probability of the V vectors in the discrete latent space;
program code to predict a future point cloud sequence according to the inferred future latent space and the past point cloud sequences; and
program code to denoising the predicted, future point cloud sequences using a denoising diffusion probabilistic model (DDPM).
10 . The non-transitory computer-readable medium of claim 9 , in which the program code to train comprises program code to feed a training encoder of the VQ-CVAE framework with a future point cloud, the input previously sampled latent space, and past point cloud sequences to predict a future latent space.
11 . The non-transitory computer-readable medium of claim 9 , in which the program code to output comprises:
program code to feed an inference encoder of the trained VQ-CVAE with the previously sampled latent space, and past point cloud sequences;
program code to infer, by the inference encoder, a classification over quantized vectors; and
program code to sample the future latent space sampled from the output categorical distribution.
12 . The non-transitory computer-readable medium of claim 9 , in which the program code to sample comprises program code to sample, by a decoder, the future latent space from the output categorical distribution.
13 . The non-transitory computer-readable medium of claim 9 , in which the program code to predict comprises program code to predict, by a decoder, the future point cloud at time t in response to the sampled latent space and features of past point cloud sequences.
14 . The non-transitory computer-readable medium of claim 9 , in which the program code to denoise comprises:
program code to perform a partial denoising process on the predicted, future point cloud sequence to generate a denoised, future point cloud sequence; and
program code to perform a partial diffusion process on the denoised, future point cloud sequence.
15 . The non-transitory computer-readable medium of claim 14 , in which the program code to perform the partial denoising process comprises:
program code to add noise to a point cloud sequence sample including the predicted, future point cloud sequence and a previously predicted point cloud sequence; and
program code to remove the noise from the point cloud sequence sample over a predetermined number of steps to provide the denoised, future point cloud sequence.
16 . The non-transitory computer-readable medium of claim 9 , further comprising program code to plan a trajectory of an ego vehicle according to a predicted, future point cloud sequence of agents within a scene surrounding the ego vehicle.
17 . A system for sequential point cloud forecasting, the system comprising:
a vector-quantized (VQ) conditional variational autoencoder (VQ-CVAE) training module to train a VQ-CVAE framework to map an output to a closest vector in a discrete latent space to obtain a future latent space;
a trained VQ-CVAE to output a categorical distribution of a probability of V vectors in the discrete latent space in response to an input previously sampled latent space and past point cloud sequences;
a latent space inference model to sample an inferred future latent space from the categorical distribution of a probability of the V vectors in the discrete latent space;
a point cloud prediction model to predict a future point cloud sequence according to the inferred future latent space and the past point cloud sequences; and
a point cloud denoising model to denoise the predicted, future point cloud sequences using a denoising diffusion probabilistic model (DDPM).
18 . The system of claim 17 , in which the VQ-CVAE) training module is further to feed a training encoder of the VQ-CVAE framework with a future point cloud, the input previously sampled latent space, and past point cloud sequences to predict a future latent space.
19 . The system of claim 17 , in which the point cloud denoising model is further to perform a partial denoising process on the predicted, future point cloud sequence to generate a denoised, future point cloud sequence, and to perform a partial diffusion process on the denoised, future point cloud sequence.
20 . The system of claim 17 , further comprising a planner module to plan a trajectory of an ego vehicle according to a predicted, future point cloud sequence of agents within a scene surrounding the ego vehicle.