Skip to main content

samples_dataframe

edgefirst_client

Function samples_dataframe 

Source 
pub fn samples_dataframe(samples: &[Sample]) -> Result<DataFrame, Error>
Expand description

Create a DataFrame from a slice of samples with the 2026.04 schema.

Each annotation in each sample becomes one row. Columns where every value is null are automatically dropped, so the result only contains columns that carry data. The name column is always present.

§Schema (2026.04)

  • name: Sample name (String) - ALWAYS PRESENT
  • frame: Frame number (UInt32)
  • object_id: Object tracking ID (String)
  • label: Object label (Categorical)
  • label_index: Label index (UInt64)
  • group: Dataset group (Categorical)
  • polygon: Segmentation polygon rings (List<List>)
  • box2d: 2D bounding box [cx, cy, w, h] (Array<Float32, 4>)
  • box3d: 3D bounding box [x, y, z, w, h, l] (Array<Float32, 6>)
  • mask: PNG-encoded raster mask (Binary)
  • box2d_score: Box2d confidence (Float32)
  • box3d_score: Box3d confidence (Float32)
  • polygon_score: Polygon confidence (Float32)
  • mask_score: Mask confidence (Float32)
  • size: Image size [width, height] (Array<UInt32, 2>)
  • location: GPS [lat, lon] (Array<Float32, 2>)
  • pose: IMU [yaw, pitch, roll] (Array<Float32, 3>)
  • degradation: Image degradation (String)
  • iscrowd: COCO crowd flag (Boolean)
  • category_frequency: LVIS frequency group (Categorical)
  • neg_label_indices: Verified-absent label indices (List)
  • not_exhaustive_label_indices: Incomplete label indices (List)
  • timing: Pipeline timing (Struct{load, preprocess, inference, decode} of Int64)

§Example

use edgefirst_client::{Client, samples_dataframe};

let samples = client
    .samples(dataset_id, Some(annotation_set_id), &[], &[], &[], None)
    .await?;
let df = samples_dataframe(&samples)?;
println!("DataFrame shape: {:?}", df.shape());