anuragsahu/3d-point-cloud-object-reconstruction-from-monoclar-rgb-image

Implementation for ``A Point Set Generation Network for 3D Object Reconstruction from a Single Image''

CV Project Page

Paper Title

37, A Point Set Generation Network for 3D Object Reconstruction from a Single Image. Link to paper

Participants

• Anurag Sahu (2018121004)
• Apoorva Srivastava (2019702014)
• Kajal Sanklecha (2019801006)

Quick Output

Output -    Input             Output V1              Output V2.

Instructions to replicate this on Local Linux System

• make sure you have git installed on your system

• make sure that you have python 2 installed on your system
  Steps:

• Get the path for python2 which python2 this will return the path where your python2 is installed

• Create a virtual Environment with python 2

example : virtualenv -p usr/bin/python2 testVenv

• clone the git repo and get into the repo

git clone https://github.com/AnuragSahu/CV_Project_2020_3D_Object_Reconstruction_from_a_Single_Image.git
cd ./CV_Project_2020_3D_Object_Reconstruction_from_a_Single_Image/

• Activate the environment

source ../testVenv/bin/activate

• install the requirements

pip install -r ./requirements.txt

• to test
  download the weights file : Link

cd ./V1/src/
python vanillaReconstruct.py \
../images/1.png \
../images/1_m.png \
../vanillaVersionWeights.pkl

python viz.py ../images/1.txt

Test on your Image

make sure that you have the mask image which filters out your object of interest from the background and also the image and mask are of size 256*192

• Place the image and its mask to images folder (lets say that image has name myImage.png and its mask image is myImageMask.png)
• run the code to get the 3d point clouds

python vanillaReconstruct.py \
../images/myImage.png \
../images/myImagemask.png \
../vanillaVersionWeights.pkl

• Visualize the point cloud

python viz.py \
../images/myImage.txt

3d-point-cloud-object-reconstruction-from-monoclar-rgb-image's Issues

[Theory doubt] What is the Random Vector?

Some boubts needs to be leared about the Random Vector Described in the Paper?

• What is the use of Random Vector?
• Where is the Random Vector comming from?
• What is the size of the random vector?

Presentation And Final Report.

Create the Presentation and Final Report for the project.

Train Two Prediction Version

This Might take a while but it should be done within 5 Days, the mini tasks to be accomplished are

1. Get the Training Code (#2)
2. Select the Classes (Approx 10) to train the data on
3. Get the Respective Dateset from the Shape-net
4. Train the Network and report the .log file
5. Also get the Training Loss Curve and Validation Loss Curve

Link to the weight is invalid

Hi, thank you very much for your work. It's so great. However, while downloading the model weights for testing, I found that the link is dead. Could you please provide a new link, if it is convenient for you. Thank you very much.

Code for Two Prediction Branch Version.

The code for Two Prediction Branch Version needs to be Uploaded by Today, the following files are expected to be Uploaded:

• trainTwoBranch.py
• twoBranchReconstruct.py
• All other files to should also be uploaded from vanilla version.
  • viz.py
  • tf_nndistance_so.so
  • BatchFetcher.py