Hello, I have benefited greatly from this open source project. Thank you for this elegant code.
(1 ) I found that the following code (in yolo_tensorflow/test.py) may be not rigorous.
The upper code is used to obtain the index of class (which has the highest class probability) for each predicted bounding box that class-specific confidence score >= 0.2
(2) During detection, I print(np.transpose(filter_mat_boxes))
(3) Then i print (probs_filtered)
(4) Finally i print (classes_num_filtered), but i found the result is
For the bounding box (4, 3, 0) and (4, 3, 1), the class index apparently should be 14 based on probs_filtered
In a word, i think the "classes_num_filtered" should be modified to:
After modification, I print (classes_num_filtered), the result is:
For example, we want to detect the following picture by YOLO
If we detect the picture without modification, the result is:
But if we detect the same picture with modification, the result is:
Maybe my comprehension is wrong. Could you give me a reasonable interpretation. Thank you in advance~ @hizhangp
Hello, I have benefited greatly from this open source project. Thank you for this elegant code.
(1 ) I found that the following code (in yolo_tensorflow/test.py) may be not rigorous.
(2) During detection, I print(np.transpose(filter_mat_boxes))
(3) Then i print (probs_filtered)
(4) Finally i print (classes_num_filtered), but i found the result is
For the bounding box (4, 3, 0) and (4, 3, 1), the class index apparently should be 14 based on probs_filtered
In a word, i think the "classes_num_filtered" should be modified to:
After modification, I print (classes_num_filtered), the result is:
For example, we want to detect the following picture by YOLO
If we detect the picture without modification, the result is:
But if we detect the same picture with modification, the result is:
Maybe my comprehension is wrong. Could you give me a reasonable interpretation. Thank you in advance~ @hizhangp