Artem Ignatev

please, what parameters are needed for optimal forecast of the model in BatchGenerator train_dataset.generate(batch_size=batch_size, shuffle=True, train=True, ssd_box_encoder=ssd_box_encoder, convert_to_3_channels=True, equalize=True, brightness=(0.5, 2, 0.5), flip=0.5, translate=False, scale=False, max_crop_and_resize=(img_height, img_width, 1, 3), #...

when loading my training weights the same tensor is predicted model.load_weights('/home/artem/object detection/virtual/Mobilenet-ssd-keras/MobilenetWeights/mobilenet.h5', by_name=True,skip_mismatch=True) y_pred = model.predict(input_images) print(y_pred) [[[0.25 0.25 0.25 ... 0.1 0.2 0.2 ] [0.25 0.25 0.25 ... 0.1...

in file: infer_mobilenet_ssd.py y_pred_decoded = decode_y(y_pred, confidence_thresh=0.25, iou_threshold=0.45, top_k=100, input_coords='centroids', normalize_coords=True, img_height=img_height, img_width=img_width) print(y_pred_decoded) out: [ [ ] ]