Getting Negative Bounding Boxes Values

[robertokcanale]

Hi, I am training my own Keras model to use YOLO3 for inference, but i get negative and wierd bounding boxes, if any at all. I am trying to segment a hand-pressure image into PALM THUMB FINGERS (0,1,2 respectively for the classes). I follow all the steps for custom training (with a small dataset of 200 images) but i get the following error: Found 1 boxes for img [ -49.221752 -130.95093 -45.08555 188.86354 ] 0.31854975

Or better, I get negative positions for my bounding boxes, why is that?

I attach a photo of my hand, its annotation in .txt, the whole train file for my script, my classes, and the image I get out of it. Some of the code was taken from: [https://pylessons.com/YOLOv3-custom-training/]

1.txt my_classes.txt train.txt

[Anyman552]

How do you convert from darknet-format to the format of keras-yolo3 (x_min,y_min,x_max,y_max,class_id) ? Have you checked if this conversion doesn't mess up anything? In the beginning I had similar problems because of that reason.

[robertokcanale]

This is what i get for image 11 in labelImg and after my conversion. I use the following script for converting ![

Screenshot from 2021-03-15 11-07-02](https://user-images.githubusercontent.com/58591956/111137107-b68d0200-857e-11eb-83f2-b5d5d60b22ee.png)

[robertokcanale]

import os import cv2 #changed for multiple classes def prep_train_txt(): width = 362 height = 247 hands = os.path.join(os.getcwd(),"hands") file_object = open('train.txt', 'w')
i=0 for img in os.listdir(hands): if img[-3:] == 'txt': nameee = img[:-3] + "png" img_name = os.path.join("hands" , nameee) img_name = cv2.imread(img_name) file_object.write("hands/" + nameee + " " ) print(img) #initialize text to write text=[] with open(os.path.join(hands,img)) as f: lines = f.readlines() print(len(text), type(text), text) #print(lines[0], type(lines)) for i in range(len(lines)): ans = lines[i] line = ans.split(" ") #print(line[4], type(line)) #Getting the x and y position of the bounding boxes, assuming these computations are correct, im not so sure x1 , y1 , w , h = float(line[1]),float(line[2]),float(line[3]) ,float(line[4][:-1]) x1 , y1 , w , h = x1 - w/2 , y1-h/2 , w , h x1 , y1 , w , h = int(x1width), int(y1height), int(wwidth),int(hheight) text.append(str(x1) + "," + str(y1) + "," + str(x1+w) + "," + str(y1+h) + "," + str(int(line[0])) + " ")
file_object.write(text[i])

            file_object.write("\n")

prep_train_txt()

[robertokcanale]

I also get my BB from https://github.com/pythonlessons/YOLOv3-object-detection-tutorial/tree/master/YOLOv3-custom-training

[Anyman552]

The coordinates are looking good. Do you provide the right path to the annotation_path (https://github.com/qqwweee/keras-yolo3/blob/e6598d13c703029b2686bc2eb8d5c09badf42992/train.py#L17) ?

Have you edited yolov3.cfg to your needs? Look at : https://github.com/AlexeyAB/darknet#how-to-train-to-detect-your-custom-objects

However, the code is working for me. I use a bigger dataset, but I don't think, that this should matter.

[Wsine]

@robertokcanale I meet this problem today. have you figure where is the problem?

[robertokcanale]

Hi, no, I just switched to yolov5. I highly suggest to change.

On Thu, 14 Oct 2021, 11:19 Jankin Wei, @.***> wrote:

@robertokcanale https://github.com/robertokcanale I meet this problem today. have you figure where is the problem?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/qqwweee/keras-yolo3/issues/747#issuecomment-943173236, or unsubscribe https://github.com/notifications/unsubscribe-auth/AN7AVVC7DGFGXRJEZ5QJJVTUG2OAZANCNFSM4ZCTIKHQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.