create SLP project from scratch using python

[mmmahdiii]

I have keypoint of each node for each frame of a video, I need a code like below to create an SLEAP project.

import sleap
from sleap import Labels, Video, LabeledFrame, Instance, Skeleton
from sleap.skeleton import Node

# Create a new SLEAP project
labels = Labels()

# Create a video object and add it to the project
video = Video.from_filename("video.mp4")
labels.add_video(video)

# Create a labeled frame and add it to the project
frame = LabeledFrame(video=video, frame_idx=0)
labels.frames.append(frame)

# Create a skeleton and add it to the project
skeleton = Skeleton()
skeleton.add_node(Node("head"))
skeleton.add_node(Node("tail"))
labels.add_skeleton(skeleton)

# Create an instance and add it to the labeled frame
instance = Instance(skeleton=skeleton)
instance.add_node("head", (100, 100))
instance.add_node("tail", (200, 200))
frame.add_instance(instance)

# Save the project to a SLEAP file
sleap.save_file(labels, "labels.slp")

[roomrys]

Hi @mmmahdiii,

When creating a Labels object, create it from the inside out by creating object that are referenced by other objects first:

import sleap
from sleap import Labels, Video, LabeledFrame, Instance, Skeleton
from sleap.instance import Point
from sleap.skeleton import Node

# Create a video object
video_path = r"path/to/video.mp4"
video = Video.from_filename(video_path)

# Create a skeleton
nodes = ["head", "tail"]
skeleton = Skeleton(name="my_skeleton")
skeleton.add_nodes(nodes)
skeleton.add_edge("head", "tail")

# Initialize a list of labeled frames
lfs = []

# 0. Create a list of instances for this labeled frame
instances = []

# 1. Create an instance and append to list of instances
points = dict(head=Point(100, 100), tail=Point(200, 200))
inst = Instance(skeleton=skeleton, points=points)
instances.append(inst)

# 2. Create a labeled frame and add it to the list
frame = LabeledFrame(video=video, frame_idx=0, instances=instances)
lfs.append(frame)

# Repeat steps 0-2 for all frames you would like to add (preferably in a for-loop)
...

# Create the Labels object and save to an slp
ds = r"path/to/save/labels.slp"
labels = Labels(lfs)
labels.save(ds)

Thanks,
Liezl

[isabellow]

Hi @roomrys,

I'm using this approach to make a SLEAP project using labeled frames from different GUI. I think it's largely working well--thank you!

I'm running into issues with the video object. My labeling GUI outputs video frames as a numpy array called images of shape (num_frames, width, height, 1). So I am using video = Video.from_numpy(images) to create the video object (instead of Video.from_filename(video_path) as above). This seems to work fine for labeling the frames (or at least it is not throwing any errors). But when I try to open the project in the labeling GUI, it says that the raw video data cannot be found.

What is the best way to save my video to the SLEAP project? Please let me know if you need more information or code snippets.

Thanks,
Isabel

[talmo]

Hi @isabellow,

Numpy videos are not very well supported since we don't store them on disk or in the SLP file. Do you mind opening a new discussion so we can follow up there if you're still running into issues?

In general, our standalone package sleap-io is also great for doing things like programmatically creating SLEAP projects.

Cheers,

Talmo

[isabellow]

Hi Talmo,

Thanks for the reply! I realized that Video.from_filename(video_path) will accept a path to a numpy array. From there, everything worked great. I have some preliminary tracking data that's not bad so far!

Best,
Isabel

[mmmahdiii]

Hi @roomrys,
it works perfectly. You always helped me.

Thanks a million!
Mahdi