Taking a uniform random variable as the value of RGB may generate a large amount of gray, white, and black, which are often not the colors we want.
The cv::applyColorMap can easily generate a random RGB palette, and you can choose a favorite color map from the list here
Example for C++11:
#include <algorithm>
#include <numeric>
#include <random>
#include <opencv2/opencv.hpp>
std::random_device rd;
std::default_random_engine re(rd());
// Generating randomized palette
cv::Mat palette(1, 255, CV_8U);
std::iota(palette.data, palette.data + 255, 0);
std::shuffle(palette.data, palette.data + 255, re);
cv::applyColorMap(palette, palette, cv::COLORMAP_JET);
// ...
// Picking random color from palette and drawing
auto randColor = palette.at<cv::Vec3b>(i % palette.cols);
cv::rectangle(img, cv::Rect(0, 0, 100, 100), randColor, -1);
Example for Python3:
import numpy as np, cv2
palette = np.arange(0, 255, dtype=np.uint8).reshape(1, 255, 1)
palette = cv2.applyColorMap(palette, cv2.COLORMAP_JET).squeeze(0)
np.random.shuffle(palette)
# ...
rand_color = tuple(palette[i % palette.shape[0]].tolist())
cv2.rectangle(img, (0, 0), (100, 100), rand_color, -1)
If you don't need so many colors, you can just cut the palette to the desired length.