TO_CLEAN.txt

import torch
boxes = torch.ones(7,4)
def object_detection_api(img_path, threshold=0.5, rect_th=3, text_size=3, text_th=3):



Bajar el tiempo de deteccion.
Minimizar lo más posible el tiempo de ejecucion.
Ser lo más pythonic posible.

Transforms in images
https://discuss.pytorch.org/t/pytorch-pil-to-tensor-and-vice-versa/6312


Improve visualization non real time.

Make random colors for each mask
.detach()?
Augmenting images -> https://pytorch.org/vision/stable/transforms.html


    ####################################
    # Visualizing bounding boxes
    ####################################
    from torchvision.models.detection import fasterrcnn_resnet50_fpn, FasterRCNN_ResNet50_FPN_Weights

    weights = FasterRCNN_ResNet50_FPN_Weights.DEFAULT
    transforms = weights.transforms()

    images_test_list = [transforms(d) for d in img_list]
    model = fasterrcnn_resnet50_fpn(weights=weights, progress=False)
    model = model.eval()

    outputs_pred = model(images_test_list)  # get a list of images with the same size, in this case only one image
    print(f'outputs==>>{outputs_pred}')

    detected_output = outputs_pred[0]
    detected_bbox = detected_output['boxes']

    print("For the image analysed, the following instances were detected:")
    print([weights.meta["categories"][label] for label in detected_output['labels']])
    label_str = [weights.meta["categories"][label] for label in detected_output['labels']]

    pred_classes = [COCO_INSTANCE_CATEGORY_NAMES[i] for i in outputs[0]['labels'].cpu().numpy()]
    ####################################

    ####################################
    #
    ####################################
    score_threshold = .8
    detected_objects_with_boxes = [
        draw_bounding_boxes(image_int, boxes=outputs_pred['boxes'][outputs_pred['scores'] > score_threshold], width=4)
        for image_int, outputs_pred in zip(img_list, outputs_pred)
    ]
    show(detected_objects_with_boxes)
    ####################################



predictions_model[0]['scores'].detach().cpu().numpy()
final_masks = masks_filtered > 0.5  # ?
final_masks = final_masks.squeeze(1)  # ?
Image.fromarray(merged_masks.mul(255).byte().cpu().numpy())