pangyuteng

class ConcatSpatialCoordinate(Layer):
    def __init__(self, **kwargs):
        """Concatenates the (x, y) coordinate normalised to 0-1 to each spatial location in the image.
        Allows a network to learn spatial bias. Has been explored in at least one paper,
        "An Intriguing Failing of Convolutional Neural Networks and the CoordConv Solution"
        https://arxiv.org/abs/1807.03247
        
        Improves performance where spatial bias is appropriate.
        
        Works with dynamic shapes.

pangyuteng

Using SSH through airplane WiFi that blocks port 22 (via GCP and Ubuntu)

SUMMARY

Open WIFI often allow only traffic via http and https ports. This gist contains steps to allow you to ssh to a machine by first creating a VM, switching the ssh port to listen to port 80. So now, you can ssh (port 80) into the VM, in order to ssh (port 22) into your desired machine.

original gist by guillochon, this Fork contains reformatted steps, and I switched OS from CentOs to Ubuntu... just as an exercise to get famaliarized with GCP.

https://gist.github.com/guillochon/eeaa54b328952d260472c14c559f698a

Estimated time to complete the below steps is 5 to 10 minutes (assuming the internet connection is good, and GCP is operating smoothly).

pangyuteng

cmake_minimum_required(VERSION 3.5)

project(_hello)

# Find python and Boost - both are required dependencies
find_package(PythonLibs REQUIRED)
find_package(PythonInterp REQUIRED)
find_package(PythonExtensions REQUIRED)
find_package(Cython REQUIRED)
find_package(ZLIB REQUIRED)

pangyuteng

cmake_minimum_required(VERSION 3.5)

# Find python and Boost - both are required dependencies
find_package(PythonLibs 2.7 REQUIRED)
find_package(Boost COMPONENTS python REQUIRED)

# Without this, any build libraries automatically have names "lib{x}.so"
set(CMAKE_SHARED_MODULE_PREFIX "")

# Add a shared module - modules are intended to be imported at runtime.

pangyuteng

#!/usr/bin/env/python
# 
# More of a reference of using jinaj2 without actual template files.
# This is great for a simple output transformation to standard out.
#
# Of course you will need to "sudo pip install jinja2" first!
#
# I like to refer to the following to remember how to use jinja2 :)
# http://jinja.pocoo.org/docs/templates/
#

pangyuteng

#
#   ref https://github.com/tebeka/pythonwise/blob/master/docker-miniconda/Dockerfile
#
#   miniconda vers: http://repo.continuum.io/miniconda
#   sample variations:
#     Miniconda3-latest-Linux-armv7l.sh
#     Miniconda3-latest-Linux-x86_64.sh
#     Miniconda3-py38_4.10.3-Linux-x86_64.sh
#     Miniconda3-py37_4.10.3-Linux-x86_64.sh
#

pangyuteng

use the share option in CVAT per davidblom cvat-ai/cvat#203

update CVAT docker-compose.yml to follow the below, where /home/david/shared_data is the local share.

version: "2.3"

services:
  cvat:
 volumes:

pangyuteng

work-in-progress data source abstraction architecture/design pattern

#### Data source abtraction layer

class FileStore
    def write
    def read
    def remove

pangyuteng

Sample code to run DAG jobs with python asyncio and aiohttp libraries.

example DAG

    D______
A _/  C    \
   \_/ \_E__\_F
     \ /

B

pangyuteng

import sys,os
import tensorflow as tf
import numpy as np
from tensorflow.python.estimator.model_fn import ModeKeys as Modes
tf.logging.set_verbosity(tf.logging.INFO)

w = 512
h = 512
d = 300
c = 11