Section outline

  • The quantity of images generated of this type is proportional to the size of your dataset. These files provide visual feedback on the model's behavior at different stages :

    • 'train_batchx.jpg' : illustrates the training samples, i.e., the annotated images ("ground truth") submitted to the model during the learning phase.
    • 'val_batchx_labels.jpg' : presents the images from the validation set accompanied by their actual annotations, serving as a reference for evaluation.
    • 'val_batchx_pred.jpg' : displays the predictions made by the model on the validation images. The overlay of predicted bounding boxes and identified classes allows for a visual assessment of the model's accuracy (localization and classification) and helps identify potential errors.