Bounding Box generation Error about zpy

[franferraz98]

Well yeah, it was my fault. That i['id'] == a['id'] should be i['id'] == a['image_id'].

Sigh. Sorry, had my head somewhere else when I wrote this. Thanks for your support!

[franferraz98]

The rest of the code, for if it's needed:

import logging
import math
import random
from pathlib import Path
import random
import numpy as np
 
import bpy
import zpy

from mathutils.bvhtree import BVHTree


log = logging.getLogger("zpy")


def rotation_matrix(axis, theta):
    """
    Return the rotation matrix associated with counterclockwise rotation about
    the given axis by theta radians.
    """
    axis = np.asarray(axis)
    axis = axis / math.sqrt(np.dot(axis, axis))
    a = math.cos(theta / 2.0)
    b, c, d = -axis * math.sin(theta / 2.0)
    aa, bb, cc, dd = a * a, b * b, c * c, d * d
    bc, ad, ac, ab, bd, cd = b * c, a * d, a * c, a * b, b * d, c * d
    return np.array([[aa + bb - cc - dd, 2 * (bc + ad), 2 * (bd - ac)],
                     [2 * (bc - ad), aa + cc - bb - dd, 2 * (cd + ab)],
                     [2 * (bd + ac), 2 * (cd - ab), aa + dd - bb - cc]])


def rotate(point, angle_degrees, axis=(0,1,0)):
    theta_degrees = angle_degrees
    theta_radians = math.radians(theta_degrees)
    rotated_point = np.dot(rotation_matrix(axis, theta_radians), point)
    return rotated_point

[hu-po]

That is a weird result for sure. Code seems to looks fine. What do the depth and segmentation images look like for those weird bounding boxes?

[franferraz98]

I deleted those ones, but here are some new ones:

---

In general they look fine, I wouldn't say that it has to do with the model.

[hu-po]

Interesting. Are you using your own visualization tools to show the bounding box over the image? There are a couple different styles of bounding boxes, zpy uses (x, y, width, height) as seen here. Could be that your visualization tool uses something else like (x1, y1, x2, y2)

[franferraz98]

Yes, I use my own script, but I'd say I got the style you say. Here is my script:

import json
import argparse
import cv2
from matplotlib import pyplot as plt

# Initiate argument parser
parser = argparse.ArgumentParser(
    description="Sample TensorFlow COCO-to-TFRecord converter")
parser.add_argument("-a",
                    "--ann_file",
                    help="Path to the folder where the input .coco.json files are stored.",
                    type=str)
parser.add_argument("-i",
                    "--img_dir",
                    help="Path to the folder where the input image files are stored. "
                         "Defaults to the same directory as ANN_FILE.",
                    type=str, default=None)

args = parser.parse_args()

if args.img_dir is None:
    args.img_dir = args.ann_file

img_dir=args.img_dir
annotations_file=args.ann_file

# Remove annotations
# Read JSON for annotations
d = {}
with open(args.ann_file) as f:  
    d = json.load(f)

f.close()

# Get images and annotations
images = d['images']
annotations = d['annotations']

for i in images:
    print("PATH: ", args.img_dir + i['file_name'])
    img = cv2.imread(args.img_dir + i['file_name'])
    for a in annotations:
        if i['id'] == a['id']:  # Annotation for the image
            [x,y,w,h] = a['bbox']
            start = (int(x), int(y))
            end = (int(x+w), int(y+h))
            color = (255,0,0)
            thick = 2
            cv2.rectangle(img, start, end, color, thick)
            cv2.imshow(' ',img)
            cv2.waitKey(0)
            cv2.destroyAllWindows()

[hu-po]

All good! Glad I could help.