Kyle Zhou
Results
2
comments of
Kyle Zhou
我训练多轮之后，单张图片做测试，有框，但是是乱的，我发现bounding_box.py第23行附近 area = h * w pred = tf.reshape(feature_map, shape=(-1, ANCHOR_NUM_EACH_SCALE * area, CATEGORY_NUM + 5)) pred = tf.nn.sigmoid(pred) #????? tx_ty, tw_th, confidence, class_prob = tf.split(pred, num_or_size_splits=[2, 2, 1, CATEGORY_NUM],...
I come across the same problem.