Neural analysis and video classification
Tensor flow
Deep learning framework
Cuda
Nvidia inference
CUDNN
nvidia inference
Tensor RT
Nvidia inference
Openvino
intelligent inference
- Data collection: Collect datasets relevant to training.
- Data preprocessing: Clean, normalize, and split data into training, validation, and test sets.
- Model architecture selection: Design the structure of the neural network, including layers, activations, and connections.
- Initialization: Randomly initialize weights and biases.
- Direct propagation: Passes input data through the network to generate predictions.
- CLoss calculation: Calculate the error between predictions and actuals using a chosen loss function.
- Backpropagation: Calculate loss gradients with respect to weights and biases.
- Gradient descent: Update weights and biases to minimize loss.
- Validation: Evaluate model performance on a separate validation dataset to avoid overfitting
- Hyperparameter tuning: Adjust learning speed, batch size, and other parameters to optimize performance.
- Test: Evaluate the final model on unseen test data to estimate real-world performance.
- Inference: Deploy the trained model to make predictions on new, unseen data.
Mixed approach for high performance - video analysis:
MATHEMATICS + GEOMETRIC + NEURAL
Vigilate video analysis exploits both traditional mathematical and deep-learning neural technologies, creating a mix that brings together the best of the two approaches.
The Vigilate video analysis reports any intruding objects in the video-surveillance areas and, if these are comparable to known classes such as people, vehicles, animals, smoke, etc. automatically recognizes them by classifying them; if, however, the objects in question cannot be traced back to any of the known classes, they are reported as requiring verification by the operators who can understand whether they are disguised intruders who could easily deceive recognition based solely on deep learning technologies
The new trend of “Just ai” can fail
EXCLUSIVE USE OF DEEP LEARNING NEURAL ONLY FINDS WHAT IT KNOWS
AND HE DOESN’T SEE WHAT HE DOESN’T KNOW
DEEPLEARNING
NO KNOWN ITEMS FOUND
Purely neural analysis does not detect things it does not know, in this case the covered man is not identified
The vigilate method
NOT ONLY ON DE EP LEARNING BUT ALSO A HIGH PERFORMING TRADITIONAL ANALYSIS RESULT OF RESEARCH AND DEVELOPMENT
NO KNOWN ITEMS FOUND
With mathematical and geometric analysis the presence of an unknown object is detected
