Supported File Formats
  • Dark
    Light
  • PDF

Supported File Formats

  • Dark
    Light
  • PDF

Data file formats

  • Images: jpeg, jpg, png, webp, tiff, tif, bmp, jfif
  • Videos:  webm, mp4, avi, x-m4v, 3gpp, x-matroska (mkv), x-msvideo
  • We recommend uploading videos in WebM format for optimal performance. See full explanation here.
  • As long as the video you upload is under 1GB, Dataloop will convert your video to WebM format for optimal performance. Files larger than 1GB will not be converted to WebM and will be uploaded in their original format (read about Manually Converting Large Files to WebM). 
  • To ensure frame-accurate annotation in training your model or for other uses, you will need to export your video file’s JSON from the Dataloop platform and use the WEBM modality URL to stream the file, since annotations are frame-accurate only with respect to that WEBM file.
  • For Audio: mp3, ogg, wav
  • JSON files (also used for applications such as Modalities)

Annotations Formats

Items have a collection of annotations attached to them(either generated in Dataloop or uploaded via SDK). Both items and annotations are searchable in the Dataset browser via standard properties or Metadata values.

  • Download annotations -

  • Upload Annotations -

    • Annotations JSON can be uploaded from the Dataloop UI (annotations studio) to a specific item.

      In the annotation studio, switch to the "ÏTEM” panel and click the "Ïmport annotations” icon, then choose to paste the JSON or select a JSON file from your file system.

    • From the SDK it is possible to upload:

    Video Formats

    Important
    • We recommend uploading video files in WebM format for the best functionality and speed.
    • As long as the video you upload is under 1.07 GB, Dataloop will convert your video to WebM format for optimal performance. Files larger than 1.07 GB will not be converted to WebM and will be uploaded in their original format (read about Manually Converting Large Files to WebM).  
    • To ensure frame-accurate annotation in training your model or for other uses, you will need to export your video file’s JSON from the Dataloop platform and use the WEBM modality URL to stream the file, since annotations are frame-accurate only with respect to that WEBM file.

Data Management for Video

Dataloop supports data management for a variety of video file formats.
With data management, you can clone, delete, arrange in folders, version across multiple datasets, filter based on metadata and more...

Video Labeling

To achieve the highest labeling performance our labeling studio requires video files to be in WebM format.

Storing videos on external cloud storage (S3/GCS/Azure) may result in latencies when serving the video to annotators. For optimal video annotation performances, we recommend storing video files on the Dataloop storage.

Recommended Python WEBM Conversion

This script uses OpenCV and tqdm
Please make sure to download them using these links: OpenCV,  tqdm.
import cv2
import tqdm
import os

# input video
input_filepath = r'C:\a\b\video.mp4'

# web creation using opencv 
name, ext = os.path.splitext(input_filepath)
output_filepath = name + '.webm'
cap_reader = cv2.VideoCapture(input_filepath)
w = int(cap_reader.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(cap_reader.get(cv2.CAP_PROP_FRAME_HEIGHT))
fps = cap_reader.get(cv2.CAP_PROP_FPS)
fourcc = 'VP80'
cap_writer = cv2.VideoWriter(output_filepath,
                             cv2.VideoWriter_fourcc(*fourcc),
                             fps,
                             (w, h))
i_frame = 0
pbar = tqdm.tqdm(total=cap_reader.get(cv2.CAP_PROP_FRAME_COUNT))
while True:
    ret, frame = cap_reader.read()
    if not ret:
        break
    i_frame += 1
    pbar.update()
    cap_writer.write(frame)
cap_writer.release()