n1ckfg

Steps to build TensorFlow v1.12.0 C-library on Jetson Nano for GPU

Jetson Nano configuration

• ARM 64 (aarch64)
• gcc 7.3
• cuda 10
• cudnn 7

TensorFlow configuration

• v1.12.0

n1ckfg

import numpy as np

def perspective_fov(fov, aspect_ratio, near_plane, far_plane):
	num = 1.0 / np.tan(fov / 2.0)
	num9 = num / aspect_ratio
	return np.array([
		[num9, 0.0, 0.0, 0.0],
		[0.0, num, 0.0, 0.0],
		[0.0, 0.0, far_plane / (near_plane - far_plane), -1.0],
		[0.0, 0.0, (near_plane * far_plane) / (near_plane - far_plane), 0.0]

n1ckfg

// licensed with CC BY-NC-SA 4.0 https://creativecommons.org/licenses/by-nc-sa/4.0/
await loadScript("https://unpkg.com/latk@1.0.3/latk.js")

latk = Latk.read("https://raw.githubusercontent.com/LightningArtist/latk-test-files/main/latk_logo.latk")

p = new P5({mode: 'WEBGL'}) // loads p5js library, comment this line after using it once
p.hide() // hide p5js canvas.

counter = 0
marktime = 0

n1ckfg

import csv

with open("tiltset_credits_unique_13908_cleaned.csv", "r") as file:
	csv_reader = csv.reader(file)

	extra_comma_counter = 0

	for i, line in enumerate(csv_reader):
		num_commas = len(line) - 1
		if (num_commas > 1):

n1ckfg

import torch
import onnxruntime as ort

torch.cuda.is_available() # Nvidia or AMD GPU
torch.backends.mps.is_available() # Apple GPU

ort.get_device() # any GPU

n1ckfg

#!/bin/bash

### steps ####
# verify the system has a cuda-capable gpu
# download and install the nvidia cuda toolkit and cudnn
# setup environmental variables
# verify the installation
###

### to verify your gpu is cuda enable check

n1ckfg

import os

with open("names.txt", "r") as file:
    lines = file.readlines()
    
    for line in lines:
        folder_name = line.strip().replace(" ", "_")
        os.makedirs(folder_name, exist_ok=True)
        print(f"Created folder: {folder_name}")

n1ckfg

# https://subscription.packtpub.com/book/data/9781788474443/9/ch09lvl1sec12/restoring-a-3d-point-from-two-observations-through-triangulation

import cv2
import numpy as np

# camera projection matrices
P1 = np.eye(3, 4, dtype=np.float32)
P2 = np.eye(3, 4, dtype=np.float32)
P2[0, 3] = -1

n1ckfg

/**
 * Resize the image to a new width and height using nearest neigbor algorithm. To make the image scale
 * proportionally, use 0 as the value for the wide or high parameter.
 * For instance, to make the width of an image 150 pixels, and change
 * the height using the same proportion, use resize(150, 0). 
 * Otherwise same usage as the regular resize().
 * 
 * Note: Disproportionate resizing squashes the "pixels" from squares to rectangles. 
 * This works about 10 times slower than the regular resize. Any suggestions for performance increase are welcome.
 */

n1ckfg

/**
 * Resize the image to a new width and height using nearest neighbor algorithm.
 * To make the image scale proportionally,
 * use 0 as the value for the wide or high parameters.
 * For instance, to make the width of an image 150 pixels,
 * and change the height using the same proportion, use resize(150, 0). 
 * Otherwise same usage as the regular resize().
 * 
 * Note: Disproportionate resizing squashes "pixels" from squares to rectangles. 
 * This works about 10 times slower than the regular resize.