Installation script for facebookresearch / pifuhd

install.sh

#!/bin/bash

minVersion="$(python3 -V 2>&1 | grep -Po '(?<=Python 3\.)([0-9]+)')"

if (( "$minVersion" <= 5 )); then
    echo "You're python version is too old. You need at least Python 3.6.x" >&2
    echo "Try the following: "
    echo "    sudo add-apt-repository ppa:deadsnakes/ppa"
    echo "    sudo apt update"
    echo "    sudo apt upgrade"
    echo "    sudo apt install python3.8 python3.8-venv python3.8-dev python3.8-gdbm"
    echo "Then make sure that Python 3.8 is the default interpreter by either configuring update-alternatives or using an alias in your .bashrc."
    echo "You can find more information here: https://medium.com/analytics-vidhya/installing-python-3-8-3-66701d3db134"
    exit 1
fi

installPath="$(pwd)"
if [ -n "$1" ]; then
    mkdir -p "$1" || exit 1
    installPath="(realpath "$1")"
fi

extraPackages=(
    "liboctomap-dev"
    "libfcl-dev"
    "libspatialindex-dev"
    "libgl1-mesa-glx"
)

installPackages=()

for package in "${extraPackages[@]}"; do
    if ! dpkg-query -W -f='${Status}\n' "$package" 2>/dev/null | grep -q "install ok"; then
        installPackages+=("$package")
    fi
done

if (( "${#installPackages[@]}" > 0 )); then
    echo "The following additional packages are needed for trimesh[all], rtree and opencv2: ${installPackages[@]}"
    sudo apt install "${installPackages[@]}" || exit 1
fi

versionError=false

fclVersion="$(dpkg-query -W -f='${Version}\n' libfcl-dev)"
if ! [[ "$fclVersion" =~ ^0\.5.* ]]; then
    echo "libfcl-dev is only available in version $fclVersion but we need at least version 0.5.0." >&2
    versionError=true
fi

octoVersion="$(dpkg-query -W -f='${Version}\n' liboctomap-dev)"
if ! [[ "$octoVersion" =~ ^1\.[89].* ]]; then
    echo "liboctomap-dev is only available in version $octoVersion but we need at least version 1.8.0." >&2
    versionError=true
fi

if $versionError; then
    echo "You can install the following packages manually and restart the script:"
    echo "    http://mirrors.kernel.org/ubuntu/pool/universe/f/fcl/libfcl0.5_0.5.0-5_amd64.deb"
    echo "    http://mirrors.kernel.org/ubuntu/pool/universe/f/fcl/libfcl-dev_0.5.0-5_amd64.deb"
    echo "    http://mirrors.kernel.org/ubuntu/pool/universe/o/octomap/liboctomap1.8_1.8.1+dfsg-1_amd64.deb"
    echo "    http://mirrors.kernel.org/ubuntu/pool/universe/o/octomap/liboctomap-dev_1.8.1+dfsg-1_amd64.deb"
    echo ""
    echo "Example:"
    echo "    wget http://mirrors.kernel.org/ubuntu/pool/universe/f/fcl/libfcl0.5_0.5.0-5_amd64.deb"
    echo "    sudo dpkg -i libfcl0.5_0.5.0-5_amd64.deb"
    exit 1
fi

cd "$installPath"

if ! [ -d "pifuhd" ]; then
    git clone https://github.com/facebookresearch/pifuhd.git || exit 1
fi

cd pifuhd

if ! [ -f "checkpoints/pifuhd.pt" ]; then
    sh ./scripts/download_trained_model.sh
fi

if ! [ -d ".venv" ]; then
    python3 -m venv .venv
fi

source .venv/bin/activate

pip install --upgrade pip || exit 1

pip install numpy || exit 1

pip install torch==1.4.0+cpu torchvision==0.5.0+cpu -f https://download.pytorch.org/whl/torch_stable.html || exit 1

pip install wheel || exit 1
pip install opencv-python tqdm matplotlib scikit-image pyopengl trimesh[all] pycocotools || exit 1

mkdir -p content
cd content
if ! [ -d "lightweight-human-pose-estimation.pytorch" ]; then
    git clone https://github.com/Daniil-Osokin/lightweight-human-pose-estimation.pytorch.git
fi
cd lightweight-human-pose-estimation.pytorch
wget -nc https://download.01.org/opencv/openvino_training_extensions/models/human_pose_estimation/checkpoint_iter_370000.pth

cat > "getrect.py" << EOM
import torch
import cv2
import numpy as np
from models.with_mobilenet import PoseEstimationWithMobileNet
from modules.keypoints import extract_keypoints, group_keypoints
from modules.load_state import load_state
from modules.pose import Pose, track_poses
import demo
import argparse

parser = argparse.ArgumentParser()
parser.add_argument('-i', '--image_path', type=str, default=None)
args = parser.parse_args()

def get_rect(net, images, height_size, cpu=False):
    net = net.eval()

    stride = 8
    upsample_ratio = 4
    num_keypoints = Pose.num_kpts
    previous_poses = []
    delay = 33
    for image in images:
        rect_path = image.replace('.%s' % (image.split('.')[-1]), '_rect.txt')
        img = cv2.imread(image, cv2.IMREAD_COLOR)
        orig_img = img.copy()
        orig_img = img.copy()
        heatmaps, pafs, scale, pad = demo.infer_fast(net, img, height_size, stride, upsample_ratio, cpu=cpu)

        total_keypoints_num = 0
        all_keypoints_by_type = []
        for kpt_idx in range(num_keypoints):  # 19th for bg
            total_keypoints_num += extract_keypoints(heatmaps[:, :, kpt_idx], all_keypoints_by_type, total_keypoints_num)

        pose_entries, all_keypoints = group_keypoints(all_keypoints_by_type, pafs, demo=True)
        for kpt_id in range(all_keypoints.shape[0]):
            all_keypoints[kpt_id, 0] = (all_keypoints[kpt_id, 0] * stride / upsample_ratio - pad[1]) / scale
            all_keypoints[kpt_id, 1] = (all_keypoints[kpt_id, 1] * stride / upsample_ratio - pad[0]) / scale
        current_poses = []

        rects = []
        for n in range(len(pose_entries)):
            if len(pose_entries[n]) == 0:
                continue
            pose_keypoints = np.ones((num_keypoints, 2), dtype=np.int32) * -1
            valid_keypoints = []
            for kpt_id in range(num_keypoints):
                if pose_entries[n][kpt_id] != -1.0:  # keypoint was found
                    pose_keypoints[kpt_id, 0] = int(all_keypoints[int(pose_entries[n][kpt_id]), 0])
                    pose_keypoints[kpt_id, 1] = int(all_keypoints[int(pose_entries[n][kpt_id]), 1])
                    valid_keypoints.append([pose_keypoints[kpt_id, 0], pose_keypoints[kpt_id, 1]])
            valid_keypoints = np.array(valid_keypoints)
            
            if pose_entries[n][10] != -1.0 or pose_entries[n][13] != -1.0:
              pmin = valid_keypoints.min(0)
              pmax = valid_keypoints.max(0)

              center = (0.5 * (pmax[:2] + pmin[:2])).astype(np.int)
              radius = int(0.65 * max(pmax[0]-pmin[0], pmax[1]-pmin[1]))
            elif pose_entries[n][10] == -1.0 and pose_entries[n][13] == -1.0 and pose_entries[n][8] != -1.0 and pose_entries[n][11] != -1.0:
              # if leg is missing, use pelvis to get cropping
              center = (0.5 * (pose_keypoints[8] + pose_keypoints[11])).astype(np.int)
              radius = int(1.45*np.sqrt(((center[None,:] - valid_keypoints)**2).sum(1)).max(0))
              center[1] += int(0.05*radius)
            else:
              center = np.array([img.shape[1]//2,img.shape[0]//2])
              radius = max(img.shape[1]//2,img.shape[0]//2)

            x1 = center[0] - radius
            y1 = center[1] - radius

            rects.append([x1, y1, 2*radius, 2*radius])

        np.savetxt(rect_path, np.array(rects), fmt='%d')

net = PoseEstimationWithMobileNet()
checkpoint = torch.load('checkpoint_iter_370000.pth', map_location=torch.device('cpu'))
load_state(net, checkpoint)

get_rect(net.cpu(), [args.image_path], 512, cpu = True)
EOM

cd "${installPath}/pifuhd"

git apply - 2>/dev/null << EOM
diff --git a/apps/recon.py b/apps/recon.py
index 8b2c98f..a5f75f3 100644
--- a/apps/recon.py
+++ b/apps/recon.py
@@ -145,7 +145,7 @@ def recon(opt, use_rect=False):
     state_dict = None
     if state_dict_path is not None and os.path.exists(state_dict_path):
         print('Resuming from ', state_dict_path)
-        state_dict = torch.load(state_dict_path)    
+        state_dict = torch.load(state_dict_path, map_location=torch.device('cpu'))
         print('Warning: opt is overwritten.')
         dataroot = opt.dataroot
         resolution = opt.resolution
@@ -162,7 +162,7 @@ def recon(opt, use_rect=False):
     
     # parser.print_options(opt)
 
-    cuda = torch.device('cuda:%d' % opt.gpu_id)
+    cuda = torch.device('cpu')
 
     if use_rect:
         test_dataset = EvalDataset(opt)
EOM

cat >> ".gitignore" << EOM
content/
.vscode/
.venv/
*.pyc
results/
checkpoints/
EOM

cat > "run.sh" << EOM
#!/bin/bash

if [ -z "\$1" ]; then
    echo "Missing parameter: $0 <image>" >&2
    exit 1
fi

image="\$(realpath "\$1")"
dirImage="\$(dirname "\$image")"

source .venv/bin/activate

cd content/lightweight-human-pose-estimation.pytorch
python getrect.py -i "\$image"

cd ../..
python -m apps.simple_test -r 256 --use_rect -i "\$dirImage"
EOM

chmod +x run.sh

mv samples_images/test_keypoints.json{,.bak} 2>/dev/null

echo "You can now go to ${installPath}/pifuhd/ and execute"
echo "    run.sh samples_images/test.png"
echo "to check if it is working correctly."
echo ""
echo "Change the value in the last line in 'run.sh' to 512 if you have enough memory."
echo "You can find the results in results/pifuhd_final/recon/"

thanks that very useful!

thanks dude

i have some issue (im not good at ai/ml stuffs) , so as per my case im using an ubuntu 20 server with 16 gigs of ram and activated the virtual environment manually and all the packages are install properly though i think something is messed :)

i have some issue (im not good at ai/ml stuffs) , so as per my case im using an ubuntu 20 server with 16 gigs of ram and activated the virtual environment manually and all the packages are install properly though i think something is messed :)

Unfortunately I am also not experienced in AI stuff. I was only able to get it installed properly. I didn't use this script since I created it. But it seems that the human pose estimation got an update and the parameter demo is no longer used as you can see here: https://github.com/Daniil-Osokin/lightweight-human-pose-estimation.pytorch/blob/1590929b601535def07ead5522f05e5096c1b6ac/modules/keypoints.py#L64
It was changed already 3 years ago in this commit: Daniil-Osokin/lightweight-human-pose-estimation.pytorch@ee9e4cc

So if you are lucky you can just delete the demo=True parameter in getrect.py on line 30 and try again.