Future Event Prediction: If and When
read more
Citations
Uncertainty-based Traffic Accident Anticipation with Spatio-Temporal Relational Learning
Event Prediction in the Big Data Era: A Systematic Survey
Am I Done? Predicting Action Progress in Videos
Event Prediction in the Big Data Era: A Systematic Survey
Predicting the future from first person (egocentric) vision: A survey
References
ImageNet Classification with Deep Convolutional Neural Networks
Automatic differentiation in PyTorch
Show, Attend and Tell: Neural Image Caption Generation with Visual Attention
Show, Attend and Tell: Neural Image Caption Generation with Visual Attention
On the importance of initialization and momentum in deep learning
Related Papers (5)
Frequently Asked Questions (9)
Q2. What is the simplest approach to modelling the TTE?
The simplest approach to modelling the CDF F̂ (t|XN ) is to quantize time in discrete bins and reduce the problem to a classification one.
Q3. What is the simplest approach to predict the TTE?
The simplest approach to predict the TTE over a continuous domain is to configure the neural network to output a real number Φ(XN ) = t̂ ∈ R+ that approximates the TTE directly (following the convention that, if the event does not occur before the time horizon ∆max, then the model outputs the value 2∆max).
Q4. What is the purpose of this paper?
Their objective in this paper introducing the problem of time-to-event prediction into computer vision by predicting future events in video before they occur.
Q5. What is the effect of the GMMH model on the prediction of traffic lights?
The authors also showed that in vehicle stopping prediction, their model outperforms an average human, which the authors contribute to the better ability of neural networks to learn domain specific priors and to capture subtle cues.
Q6. What is the loss function of the Gaussian Mixture Model?
The loss function is the negative log-likelihood of the GMM regularized by the loss already adopted for the heuristic heatmap:E(XN ,t) − log T∑ j=1 hjN (τ ;µj , σj) 〈1,h〉 + λ‖gt − h‖2 .
Q7. What is the probability of a car stopping?
The model has also learned that when approaching a green traffic light, there is a still a probability the car might stop (fig. 1 — middle row), as green might turn red before the car gets there.
Q8. What is the way to describe the TTE?
the authors describe a number of prediction models for the TTE, all implemented as neural networks Φ. These networks take as input a sequence of observations XN and output an estimate F̂ (t|XN ) of the TTE CDF.
Q9. Why do the authors only train and evaluate the prediction in the interval of (1, 5) seconds?
In contrast to the previous section, the authors only train and evaluate the prediction in the interval of (1, 5) seconds, because of generally faster pace of the action happening in the videos and to avoid issues with cuts in the TV stream.