DOC.md

TODO

[X] Print only number of people found and height
[ ] Add Elbow points to the Body_Matrix outputs and Database
[ ] Add Number of People found in Video to database

Build and Export

Change the Version number in pyproject.toml

$ python -m build 
$ twine upload dist/*

Load Models, Video and Image Frames***

Load Segmentation Model to CPU

from body_matrix import load
segment_model, segment_transforms = load.segment_model("cpu")

Load Segmentation Model to GPU

from body_matrix import load
segment_model, segment_transforms = load.segment_model("cuda")

Load Keypoints Model to CPU

from body_matrix import load
keypoints_model, keypoints_transforms = load.keypoints_model("cpu")

Load Keypoints Model to GPU

from body_matrix import load
keypoints_model, keypoints_transforms = load.keypoints_model("cuda")

Load Video

from body_matrix import load
video, frames_counts, fps, sample_frame = load.video("/Desktop/dt.mov", 90, 1)
print(frames_counts)

Load Image

from body_matrix import load
frames_path = load.image_frames("/Desktop/instagram")
print(frames_path)

Infer

Detect_Main_Target

from body_matrix import infer
from body_matrix import load

keypoints_model, keypoints_transform = load.keypoints_model("cpu")
video, frame_counts, fps, sample_frame = load.video("04_01.mp4", -90, 1)

selected_box, keypoints = infer.detect_main_target(
	sample_frame, "cpu", 0.8, keypoints_model, keypoints_transform
)

Segment_Selected_Target

from body_matrix import infer
from body_matrix import load


video, frame_counts, fps, sample_frame = load.video("04_01.mp4", -90, 1)
segment_model, segment_transform = load.segment_model("cpu")
keypoints_model, keypoints_transform = load.keypoints_model("cpu")

selected_box, keypoint = infer.detect_main_target(
	sample_frame, "cpu", 0.8, keypoints_model, keypoints_transform
)

mask, mask_image, bool_mask = infer.segment_selected_target(
	sample_frame, "cpu", selected_box, 0.99, segment_model, segment_transform
)

Filter

Keypoints_Filter

from body_matrix import load
from body_matrix import infer
from body_matrix import process

video, frame_counts, fps, sample_frame = load.video(
	"sample02.mp4", 
	-90, 
	1
)

keypoints_model, keypoints_transform = load.keypoints_model("cpu")
boxes, keypoints = infer.detect_main_target(
	sample_frame, "cpu", 0.8, keypoints_model, keypoints_transform
)

selected_kps = process.keypoints_filter(
	['nose','left_shoulder','right_shoulder'], 
	keypoints
)

print(selected_kps[nose])

Human_Segmentation_Area

from body_matrix import load
from body_matrix import infer
from body_matrix import process

video, frame_counts, fps, sample_frame = load.video(
	"sample02.mp4", 
	-90, 
	1
)

keypoints_model, keypoints_transform = load.keypoints_model("cpu")
selected_box, keypoints = infer.detect_main_target(
	sample_frame, "cpu", 0.8, keypoints_model, keypoints_transform
)

selected_kps = process.keypoints_filter(
	['nose','left_shoulder','right_shoulder'], 
	keypoints
)

segment_model, segment_transform = load.segment_model("cpu")
mask, mask_image, bool_mask = infer.segment_selected_target(
	sample_frame, "cpu", selected_box, 0.99, segment_model, segment_transform
)

segment_area = process.segmentation_area(
	sample_frame, 
	bool_mask
)

Segmentation_Contour

from body_matrix import load
from body_matrix import infer
from body_matrix import process

video, frame_counts, fps, sample_frame = load.video(
    "/content/drive/MyDrive/Body_Matrix/Raw_Vids/VID_20230102_171936~2.mp4", 
    -90, 
    2
)

keypoints_model, keypoints_transform = load.keypoints_model("cuda")
selected_box, keypoints = infer.detect_main_target(
    sample_frame, "cuda", 0.8, keypoints_model, keypoints_transform
)

selected_kps = process.keypoints_filter(
    ['nose','left_shoulder','right_shoulder'], 
    keypoints
)

segment_model, segment_transform = load.segment_model("cuda")
mask, mask_image, bool_mask = infer.segment_selected_target(
    sample_frame, "cuda", selected_box, 0.99, segment_model, segment_transform
)

segment_area = process.segmentation_area(
    sample_frame, 
    bool_mask
)

segment_contours = process.segmentation_contour(
    sample_frame, 
    bool_mask
)

Segmentation_Intersection - TO be DONE

Filter_Segmentation_Intersection - TO be DONE

Draw

Keypoint_Markers

from body_matrix import load
from body_matrix import infer
from body_matrix import process
from body_matrix import draw

video, frame_counts, fps, sample_frame = load.video(
	"sample02.mp4", 
	-90, 
	1
)

keypoints_model, keypoints_transform = load.keypoints_model("cpu")
boxes, keypoints = infer.detect_main_target(
	sample_frame, "cpu", 0.8, keypoints_model, keypoints_transform
)

selected_kps = process.keypoints_filter(
	['nose','left_shoulder','right_shoulder'], 
	keypoints
)

output = draw.keypoint_markers(
    coordinates=selected_kps,
    image=sample_frame, 
    label_font="/path/Roboto-Bold.ttf"
)

output

Circle_Label

from body_matrix import draw

middle_back = (309.86787033081055, 699.3268127441406)
updated_sample = draw.circle_label(
    image=output, 
    central_point=middle_back,
    radius=100,
    background_color = "#FF0000", 
    label = "21", 
    label_size = 100,
    label_font="/path/Roboto-Bold.ttf", 
    label_color ="#FAFF00"
)

Rectangle_Label

from body_matrix import draw
from body_matrix import load

video, frame_counts, fps, sample_frame = load.video(
    "/Users/troydo42/Desktop/TorchVision/INFER_VIDS/Nov30/00_01.mp4", 
    -90, 
    30
)

rect_sample = draw.fixed_rectangle_label(
    image=sample_frame, 
    anchor_point=[sample_frame.width/2, 100], 
    label_text="Lottee Mart",
    label_size=30, 
    label_font="/Users/troydo42/Desktop/Body_Matrixes/Roboto-Bold.ttf", 
    label_color ="#FFFFFF",
    background_color = "#11114A"
)

rect_sample

Connecting_Line

from body_matrix import draw
from body_matrix import load

video, frame_counts, fps, sample_frame = load.video(
    "/Users/troydo42/Desktop/TorchVision/INFER_VIDS/Nov30/00_01.mp4", 
    -90, 
    30
)

keypoints_model, keypoints_transform = load.keypoints_model("cpu")
boxes, keypoints = infer.detect_main_target(
    sample_frame, "cpu", 0.8, keypoints_model, keypoints_transform
)

selected_kps = process.keypoints_filter(
    [
        'left_shoulder','right_shoulder',
        'left_hip','right_hip', 
        'left_wrist','right_wrist',
        'left_ankle', 'right_ankle'
    ], 
    keypoints
)

new_sample = draw.connecting_line(
    image=rect_sample, 
    pointA=selected_kps['right_shoulder'],
    pointB=selected_kps['left_hip'],
    line_color="white", 
    line_width=12
)

new_sample

Floating Label

from body_matrix import load
from body_matrix import measure 
from body_matrix import draw
from body_matrix import infer
from body_matrix import process

video, frame_counts, fps, sample_frame = load.video(
    "/Users/troydo42/Desktop/TorchVision/INFER_VIDS/Nov30/00_01.mp4", 
    -90, 
    30
)

keypoints_model, keypoints_transform = load.keypoints_model("cpu")
boxes, keypoints = infer.detect_main_target(
    sample_frame, "cpu", 0.8, keypoints_model, keypoints_transform
)

selected_kps = process.keypoints_filter(
    [
        'left_shoulder','right_shoulder',
        'left_hip','right_hip', 
        'left_wrist','right_wrist',
        'left_ankle', 'right_ankle'
    ], 
    keypoints
)

middle_hip = measure.find_middle_point(
    selected_kps['left_hip'], 
    selected_kps['right_hip']
)

middle_shoulder = measure.find_middle_point(
    selected_kps['left_shoulder'], 
    selected_kps['right_shoulder']
)

middle_back = measure.find_middle_point(
    middle_shoulder, 
    middle_hip
)

float_sample = sample_frame
for key, value in selected_kps.items():
    print(key, value)
    float_sample = draw.floating_rectangle_label(
        image=float_sample, 
        longitude_coordinate=middle_back[0],
        point=value, 
        label_text=key, 
        label_size=16, 
        label_color="#ffffff", 
        label_font="/Users/troydo42/Desktop/Body_Matrixes/Roboto-Bold.ttf",
        background_color="#11114A"
    )    

float_sample

Draw Segmentation_Contour

from body_matrix import load
from body_matrix import infer
from body_matrix import process
from body_matrix import process
from body_matrix import draw

video, frame_counts, fps, sample_frame = load.video(
    "sample02.mp4", 
    -90, 
    1
)

keypoints_model, keypoints_transform = load.keypoints_model("cpu")
selected_box, keypoints = infer.detect_main_target(
    sample_frame, "cpu", 0.8, keypoints_model, keypoints_transform
)

selected_kps = process.keypoints_filter(
    ['nose','left_shoulder','right_shoulder'], 
    keypoints
)

segment_model, segment_transform = load.segment_model("cpu")
mask, mask_image, bool_mask = infer.segment_selected_target(
    sample_frame, "cpu", selected_box, 0.99, segment_model, segment_transform
)

segment_area = process.segmentation_area(
    sample_frame, 
    bool_mask
)

updated_contours = process.segmentation_contour(
    sample_frame, 
    bool_mask
)

contoured_image = draw.segmentation_contour(
    contour_pixels = updated_contours, 
    contour_color = "#ffffff",
    contour_size=2,
    font="/content/drive/MyDrive/Body_Matrix/Roboto-Bold.ttf", 
    image=sample_frame
)

contoured_image

Add Crown

from body_matrix import draw

crowned = draw.add_crown(
    score=1200, 
    le=selected_kps['left_ear'], 
    re=selected_kps['right_ear'], 
    top_head=top_head, 
    frame=sample_frame, 
    crown_image="/content/drive/MyDrive/Body_Matrix/crown.png",
    font_file="/content/drive/MyDrive/Body_Matrix/Roboto-Bold.ttf"
)
crowned

Draw Background Blur

from body_matrix import load, infer, process, draw

keypoints_model, keypoints_transform = load.keypoints_model("cpu")
video, frame_counts, fps, sample_frame = load.video(
    "/Users/troydo42/Desktop/Raw_Vids/VID_20230109_154951~2.mp4", 
    -90, 
    40
)

selected_box, keypoints = infer.detect_main_target(
    sample_frame, "cpu", 0.8, keypoints_model, keypoints_transform
)

blurImage = draw.blur_background_around_bbox(
    selected_box, 
    sample_frame
)

Pixelate Image

from PIL import Image
from body_matrix import load
from body_matrix import draw

video, frame_counts, fps, sample_frame = load.video(
    "/Users/troydo42/Desktop/Body_Matrixes/samples/man_03.mp4", 
    -90, 
    30
)
draw.pixelate(sample_frame, 8)

Measure

Box_Center_Coordinate

from body_matrix import load
from body_matrix import infer
from body_matrix import measure 

video, frame_counts, fps, sample_frame = load.video(
    "/Users/troydo42/Desktop/Body_Matrixes/samples/man_03.mp4", 
    -90, 
    30
)

keypoints_model, keypoints_transform = load.keypoints_model("cpu")
box, keypoints = infer.detect_main_target(
    sample_frame, "cpu", 0.8, keypoints_model, keypoints_transform
)

box

center = measure.box_center_coordinate(box)

Two_Boxes_Distance

from body_matrix import load
from body_matrix import infer
from body_matrix import measure
from torchvision.transforms.functional import to_pil_image

sample_frame_index = 30
video, frame_counts, fps, sample_frame = load.video(
    "/path/sample01.mp4", 
    -90, 
    sample_frame_index
)

keypoints_model, keypoints_transform = load.keypoints_model("cpu")
box, keypoints = infer.detect_main_target(
    frame=sample_frame, 
    device="cpu",
    min_accuracy=0.8,
    kp_model=keypoints_model, 
    kp_transforms=keypoints_transform 
)

next_frame = to_pil_image(video[sample_frame_index+1])
next_frame = next_frame.rotate(-90, expand=True)
next_box, keypoints = infer.detect_main_target(
    frame=next_frame, 
    device="cpu",
    min_accuracy=0.8,
    kp_model=keypoints_model, 
    kp_transforms=keypoints_transform 
)

moved_distance = measure.two_boxes_distance(box, next_box)
moved_distance

Box_Distance_From_Vertical_Line

from body_matrix import load
from body_matrix import infer
from body_matrix import measure
from torchvision.transforms.functional import to_pil_image

sample_frame_index = 30
video, frame_counts, fps, sample_frame = load.video(
    "/path/sample01.mp4", 
    -90, 
    sample_frame_index
)

keypoints_model, keypoints_transform = load.keypoints_model("cpu")
box, keypoints = infer.detect_main_target(
    frame=sample_frame, 
    device="cpu",
    min_accuracy=0.8,
    kp_model=keypoints_model, 
    kp_transforms=keypoints_transform 
)

distance, area = measure.distance_from_vertical_line(
    pic=sample_frame, 
    bbox=box
)

Box_Distance_From_Center

from body_matrix import load
from body_matrix import infer
from body_matrix import measure

sample_frame_index = 10
video, frame_counts, fps, sample_frame = load.video(
    "/path/sample01.mp4", 
    -90, 
    sample_frame_index
)

keypoints_model, keypoints_transform = load.keypoints_model("cpu")

box, keypoints = infer.detect_main_target(
    sample_frame, 
    device="cpu", 
    min_accuracy=0.8, 
    kp_model=keypoints_model, 
    kp_transforms=keypoints_transform
)

distance, area = measure.box_distance_from_center(sample_frame, box)

Two_Points_Distance

from body_matrix import load
from body_matrix import infer
from body_matrix import measure

sample_frame_index = 10
video, frame_counts, fps, sample_frame = load.video(
    "/path/sample01.mp4", 
    -90, 
    sample_frame_index
)

keypoints_model, keypoints_transform = load.keypoints_model("cpu")

box, keypoints = infer.detect_main_target(
    sample_frame, 
    device="cpu", 
    min_accuracy=0.8, 
    kp_model=keypoints_model, 
    kp_transforms=keypoints_transform
)

distance = measure.two_points_distance(keypoints[0], keypoints[1])
distance

Find_Middle_Point

from body_matrix import load
from body_matrix import infer
from body_matrix import measure

sample_frame_index = 10
video, frame_counts, fps, sample_frame = load.video(
    "/path/sample01.mp4", 
    -90, 
    sample_frame_index
)

keypoints_model, keypoints_transform = load.keypoints_model("cpu")

box, keypoints = infer.detect_main_target(
    sample_frame, 
    device="cpu", 
    min_accuracy=0.8, 
    kp_model=keypoints_model, 
    kp_transforms=keypoints_transform
)

x, y = measure.find_middle_point(keypoints[0], keypoints[1])
print(x.item(), y.item())

Box_Distance_From_Horizontal_Line

from body_matrix import load
from body_matrix import infer
from body_matrix import measure

sample_frame_index = 10
video, frame_counts, fps, sample_frame = load.video(
    "/path/sample01.mp4", 
    -90, 
    sample_frame_index
)

keypoints_model, keypoints_transform = load.keypoints_model("cpu")

box, keypoints = infer.detect_main_target(
    sample_frame, 
    device="cpu", 
    min_accuracy=0.8, 
    kp_model=keypoints_model, 
    kp_transforms=keypoints_transform
)

distance = measure.distance_from_horizon_line(
    sample_frame, 
    box
)
distance

Score

Two_Points_Linear_Constants

from body_matrix import load
from body_matrix import infer
from body_matrix import process
from body_matrix import score

video, frame_counts, fps, sample_frame = load.video(
    "/Users/troydo42/Desktop/Body_Matrixes/samples/man_01.mp4", 
    90, 
    1
)

keypoints_model, keypoints_transform = load.keypoints_model("cpu")
selected_box, keypoints = infer.detect_main_target(
    sample_frame, "cpu", 0.8, keypoints_model, keypoints_transform
)
selected_kps = process.keypoints_filter(
    [ 'left_ankle', 'right_ankle'], 
    keypoints
)

la = selected_kps['left_ankle']
ra = selected_kps['right_ankle']

alpha, beta = score.two_points_linear_constant(la, ra)
print(alpha, beta)

Find_Segment_Line

from body_matrix import load
from body_matrix import infer
from body_matrix import process

video, frame_counts, fps, sample_frame = load.video(
    "/Users/troydo42/Desktop/Body_Matrixes/samples/man_01.mp4", 
    90, 
    1
)

keypoints_model, keypoints_transform = load.keypoints_model("cpu")
selected_box, keypoints = infer.detect_main_target(
    sample_frame, "cpu", 0.8, keypoints_model, keypoints_transform
)
selected_kps = process.keypoints_filter(
    [
        'left_shoulder','right_shoulder', 
        'left_hip', 'right_hip',  
        'left_elbow','right_elbow',
        'left_wrist', 'right_wrist',
        'left_ankle', 'right_ankle'], 
    keypoints
)

segment_model, segment_transform = load.segment_model("cpu")
mask, mask_image, bool_mask = infer.segment_selected_target(
    sample_frame, "cpu", selected_box, 0.9, segment_model, segment_transform
)

segment_area = process.segmentation_area(
    sample_frame, 
    bool_mask
)

ls = selected_kps['left_shoulder']
rs = selected_kps['right_shoulder']

alpha, beta = two_points_linear_constant(ls, rs)
shoulder_line_coordinates = find_segment_line(segment_area, alpha, beta)

SHA_Score

from body_matrix import score

shoulder_length = 239.0083680543424
hip_length = 222.02026934494066
back_length = 287.0339701150371
leg_length = 431.9036764558864

score = score.SHA_score(
    shoulder_length, 
    hip_length, 
    back_length, 
    leg_length
)

video_SHA_Scores

### Load Video
from body_matrix import load, score
video, frames_counts, fps, sample_frame = load.video(
    "/content/drive/MyDrive/Body_Matrix/Raw_Vids/VID_20230105_174258.mp4", 
    rotate_angle=-90, 
    frame_position=1
)
print(frames_counts)
sample_frame

### Load Keypoints Model
segment_model, segment_transform = load.segment_model("cuda")
keypoints_model, keypoints_transform = load.keypoints_model("cuda")

### Iterate Over Video Frame for SHA Score
SHA_frames, SHA_scores = score.video_SHA_score(
    vid=video,
    device="cuda", 
    font_dir="/content/drive/MyDrive/Body_Matrix/Roboto-Bold.ttf",
    segment_model=segment_model,
    segment_transform=segment_transform, 
    keypoints_model=keypoints_model, 
    keypoints_transform=keypoints_transform
)

index = int(len(SHA_scores)/2)
print(SHA_scores[index])
SHA_frames[index]

Video_SHA_scores

from body_matrix import load, score
video, frames_counts, fps, sample_frame = load.video(
    "/content/drive/MyDrive/Body_Matrix/Raw_Vids/VID_20220831_161746.mp4", 
    rotate_angle=-90, 
    frame_position=1
)
print(frames_counts)
sample_frame

### Load Keypoints Model
segment_model, segment_transform = load.segment_model("cuda")
keypoints_model, keypoints_transform = load.keypoints_model("cuda")

### Iterate Over Video Frame for SHA Score
SHA_frames, SHA_scores, SHA_measures = score.video_SHA_score(
    vid=video,
    device="cuda", 
    font_dir="/content/drive/MyDrive/Body_Matrix/Roboto-Bold.ttf",
    segment_model=segment_model,
    segment_transform=segment_transform, 
    keypoints_model=keypoints_model, 
    keypoints_transform=keypoints_transform
)

index = int(len(SHA_scores)/2)
print(SHA_scores[index])
SHA_frames[index]

best_scores

import numpy as np 
import seaborn as sns

mean, median, minim, maxim = score.best_scores(SHA_scores, 1000, 1800)

find_nearest

best_frame_score, best_frame_index = score.find_nearest(SHA_scores, median)
SHA_frames[best_frame_index]

Export

Generate_Video_From_PIL_Images

from body_matrix import load, score, export

video, frames_counts, fps, sample_frame = load.video(
    "/content/drive/MyDrive/Body_Matrix/Raw_Vids/VID_20230109_114329~2.mp4", 
    rotate_angle=-90, 
    frame_position=1
)
print(frames_counts)
sample_frame

### Load Keypoints Model
segment_model, segment_transform = load.segment_model("cuda")
keypoints_model, keypoints_transform = load.keypoints_model("cuda")

### Iterate Over Video Frame for SHA Score
SHA_frames, SHA_scores = score.video_SHA_score(
    vid=video,
    device="cuda", 
    font_dir="/content/drive/MyDrive/Body_Matrix/Roboto-Bold.ttf",
    segment_model=segment_model,
    segment_transform=segment_transform, 
    keypoints_model=keypoints_model, 
    keypoints_transform=keypoints_transform
)

vid_width = sample_frame.width
vid_height = sample_frame.height
vid_output = "visualized.mp4"

export.generate_video_from_pil_images(
    pil_images=SHA_frames, 
    output=vid_output, 
    width=vid_width,
    height=vid_height
)

Generate_Seeking_Video_From_Images

import av
from body_matrix import load, export
from torchvision.transforms.functional import to_pil_image

video, frames_counts, fps, sample_frame = load.video(
    "/content/drive/MyDrive/Body_Matrix/Raw_Vids/VID_20221130_113813~2.mp4", 
    -90, 
    1
)

pil_images = []
for frame in video:
    pil_image = to_pil_image(frame)
    pil_image = pil_image.rotate(-90, expand=True)
    pil_images.append(pil_image)

export.generate_seek_video(
    pil_images=pil_images, 
    target_index= int(frames_counts/3), 
    output="fun.mp4", 
    width=sample_frame.width, 
    height=sample_frame.height
)

Generate_Instagram_Video_From_Images

import av
from body_matrix import load, export
from torchvision.transforms.functional import to_pil_image

video, frames_counts, fps, sample_frame = load.video(
    "/content/drive/MyDrive/Body_Matrix/Raw_Vids/VID_20221130_113813~2.mp4", 
    -90, 
    1
)

pil_images = []
for frame in video:
    pil_image = to_pil_image(frame)
    pil_image.rotate(-90, expand=True)
    pil_images.append(pil_image)

export.generate_instagram_vid(
    vid_name="instagram.mp4", 
    vid_width = sample_frame.width, 
    vid_height = sample_frame.height, 
    pil_images = pil_images, 
    stop_index=int(frames_counts/2), 
    fps=fps, 
    repeat_rate=2, 
    slow_motion_rate=1
)

Generate_Long_Video

import os
from body_matrix import load, export

video_dir = "/content/drive/MyDrive/Body_Matrix/Raw_Vids/"
video_list = os.listdir(video_dir)
video_links = []

for video in video_list:
    video_path = video_dir + video
    video_links.append(video_path)

pil_images = []
for vid in video_links:
    video, frames_counts, fps, sample_frame = load.video(
        vid, 
        -90, 
        1
    )
    pil_images.append(sample_frame)

export.generate_long_video(
    pil_images=pil_images, 
    scroll_speed=5,
    fps=30
)