I ran the train_frcnn.py to identify water bodies in an image using the followi

it happens when the neg_sample tuple is <code class="

Exception: 'a' cannot be empty unless no samples are taken about keras-frcnn HOT 3 OPEN

kbardool commented on July 27, 2024 2

Exception: 'a' cannot be empty unless no samples are taken

from keras-frcnn.

Comments (3)

zakrzewki commented on July 27, 2024 9

it happens when the neg_sample tuple is [[1 0 ]] only. It's not the most clever idea, but I added an exception catch for skipping such sample and continue for using another one

        if C.num_rois > 1:
            if len(pos_samples) < C.num_rois//2:
                selected_pos_samples = pos_samples.tolist()
            else:
                selected_pos_samples = np.random.choice(
                    pos_samples, C.num_rois//2, replace=False).tolist()
            try:
                selected_neg_samples = np.random.choice(
                    neg_samples, C.num_rois - len(selected_pos_samples), replace=False).tolist()
            except ValueError:
                try:
                    selected_neg_samples = np.random.choice(
                        neg_samples, C.num_rois - len(selected_pos_samples), replace=True).tolist()
                except:
                    # The neg_samples is [[1 0 ]] only, therefore there's no negative sample
                    continue

a in the error is the parameter definition used in np.random.choice: https://docs.scipy.org/doc/numpy/reference/generated/numpy.random.choice.html

Parameters: | a : 1-D array-like or intIf an ndarray, a random sample is generated from its elements. If an int, the random sample is generated as if a were np.arange(a)

from keras-frcnn.