Zong Fan zongfan2

💭

Coding is life!

zongfan2 / tic_tac_go.py

Last active May 13, 2021 11:28

Application of deep Q-learning for tic-tac-go game on PyTorch

	"""
	Modified from github repo: shakedzy/tic_tac_toe
	"""
	import numpy as np
	import torch
	import torch.nn as nn
	import os
	import time
	import random
	from collections import deque

zongfan2 / CBM3D.py

Created May 7, 2020 06:51

CBM3D algorithm for white Gaussian noise attenuation

zongfan2 / opencv_heatmap.py

Created November 26, 2019 03:02

Generate crowd heatmap with opencv and heatmap package

	import heatmap
	import cv2

	def use_heatmap(image, box_centers):
	hm = heatmap.Heatmap()
	img = hm.heatmap(box_centers, dotsize=200, size=(image.shape[1], image.shape[0]), opacity=128, area=((0, 0), (image.shape[1], image.shape[0])))
	return img

	img = "/path/to/image.jpg"
	centers = [(10, 20), (30, 40) ] # centers of heatmaps

zongfan2 / onnx_tensorrt_backend_example.py

Created November 5, 2019 10:03

ONNX model inference with onnx_tensorrt backend

	import onnx
	import argparse
	import onnx_tensorrt.backend as backend
	import numpy as np
	import time

	def main():
	parser = argparse.ArgumentParser(description="Onnx runtime engine.")
	parser.add_argument(
	"--onnx", default="/home/arkenstone/test_face_model/res50/mxnet_exported_mnet.onnx",

zongfan2 / onnx_tensorrt_inference.py

Created November 4, 2019 09:14

Acceleration inference of onnx model with TensorRT

	import tensorrt as trt
	import numpy as np
	import pycuda.autoinit
	import pycuda.driver as cuda
	import time

	model_path = "model.onnx"
	input_size = 32

	TRT_LOGGER = trt.Logger(trt.Logger.WARNING)

zongfan2 / image_client_for_trt_serving.py

Created November 1, 2019 14:34

Modified tensorrt inference server image client sample

	import argparse
	import numpy as np
	import os
	import tensorrtserver.api as tapi
	import tensorrtserver.api.model_config_pb2 as model_config
	import cv2
	import queue as q


	def model_dtype_to_np(model_dtype):

zongfan2 / deepstream_gstreamer_decoding_with_python.py

Created September 18, 2019 09:06

Use Nvidia deepstream SDK gstreamer plugin to decode H.264 video stream and output decoded frames in python

	import subprocess as sp
	import cv2
	import numpy as np

	cmd = ["gst-launch-1.0",
	"rtspsrc", "location=rtsp://admin:[email protected]/Streaming/Channels/1", "latency=100", "!",
	"queue", "!",
	"rtph264depay", "!",
	"h264parse", "!",
	"nvv4l2decoder", "drop-frame-interval=2", "!",

zongfan2 / ubuntu1804_dGPU_install_nv_deepstream.dockerfile

Last active December 22, 2021 12:46

Dockerfile to prepare DeepStream in docker for Nvidia dGPUs (including Tesla T4, GeForce GTX 1080, RTX 2080 and so on)

	From ubuntu:18.04 as base

	# install github and vim
	RUN apt-get install -y vim wget gnupg

	ENV DEBIAN_FRONTEND "noninteractive" # to skip any interactive configurations during installation
	RUN apt-get install -yq keyboard-configuration

	# install gstreamer
	RUN apt install -y \

zongfan2 / ffmpeg_python_with_gpu_acceleration.py

Created September 8, 2019 16:35

Use pipe to read ffmped decoded video frames with NVIDIA GPU hardware acceleration

zongfan2 / tf_model_post_training_quantization.py

Last active June 14, 2022 08:15

post-training quantization tensorflow model to float16

	import tensorflow as tf
	from tensorflow.core.framework import types_pb2, graph_pb2, attr_value_pb2
	from tensorflow.tools.graph_transforms import TransformGraph
	from google.protobuf import text_format
	import numpy as np

	# object detection api input and output nodes
	input_name = "image_tensor"
	output_names = ["detection_boxes", "detection_classes", "detection_scores", "num_detections"]
	# Const should be float32 in object detection api during nms (see here: https://www.tensorflow.org/api_docs/cc/class/tensorflow/ops/non-max-suppression-v4.html)