Transfer Learning avec PyTorch (torchvision / Python)
Deep Learning, convolutional neural networks. Principle of transfer learning in computer vision. How to build a hybrid model by matching a model pre-trained on large corpora (VGG19 on ImageNet) with an ad hoc classifier for binary classification (the famous cats and dogs). Performance gap between an ad hoc model and the hybrid model. Parallel with embeddings in NLP (natural language processing).
Notebook and data: http://tutoriels-data-science.blogspo...
PyTorch website: https://pytorch.org/tutorials/beginne...
Image database: https://www.microsoft.com/en-us/downl...
Course page: https://cours-machine-learning.blogsp...
00:00 Principle of transfer learning
05:58 Transfer learning with PyTorch in computer vision (VGG19)
11:09 The "Cats and Dogs" image database
13:33 Starting the notebook - Importing TRAIN images
17:18 ConvNet "from scratch" using the TRAIN database
19:40 Training the Model
21:20 Base load TEST, prediction, evaluation
22:29 Initial assessment - How to go further
25:25 Instantiation of the pre-trained VGG19 model
25:58 Applying VGG19 to an image
26:58 Prediction of one of the ImageNet classes
27:21 VGG19 architecture (features vs. classifier)
28:24 Number of layers in the dense part (classifier)
28:40 Structure of the last layer (4096 vs. 1000 ImageNet classes)
29:19 Creation of a layer specific to the problem being addressed
31:12 Grafting the new layer into VGG19
32:34 Reconfiguration of the training (loss function, optimization algorithm)
32:47 Rerunning the training (of the last layer) of the modified VGG19
35:10 Prediction on TEST and performance improvement
Auf dieser Seite können Sie das Online-Video Transfer Learning avec PyTorch (torchvision / Python) mit der Dauer stunde minuten sekunde in guter Qualität ansehen, das der Benutzer MASTER 2 SISE DATA SCIENCE 31 Mai 2024 hochgeladen hat, den Link mit Freunden und Bekannten teilen, dieses Video wurde auf Youtube bereits 788 Mal angesehen und es wurde von like den Zuschauern gefallen. Viel Spaß beim Betrachtenden Zuschauern gefallen!
