MediaPipe EfficientDet-Lite2 (Object Detector)

EfficientDet-Lite2 object detection model(s) (efficientdet_lite2_int8.tflite [7.5 MB], _float16.tflite [12.1 MB], _float32.tflite [23.1 MB]) trained on COCO (1.5M instances, 80 classes / 90 COCO label slots) and packaged with TFLITE_METADATA and embedded labels for the MediaPipe Object Detector (ObjectDetector) task and LiteRT.

Documentation: MediaPipe Object Detector Docs

Samples: Android · iOS · Python Colab · Web · Flutter

Try live: Web Demo

EfficientDet-Lite2 pairs a higher-capacity EfficientNet-Lite2 backbone with a 448 x 448 input resolution and BiFPN feature network for higher detection accuracy (especially on smaller objects) at higher compute and memory cost.

Input resolution is 448 x 448 RGB. Each .tflite embeds the 90-line COCO labels.txt directly in TFLITE_METADATA ([1, 37629, 90] class scores).

For standalone reference, this repository includes the official MediaPipe labelmap.txt (90 lines, person through toothbrush).

Files

File in Repo Root Upstream Google Path Precision Size SHA-256
efficientdet_lite2_int8.tflite object_detector/efficientdet_lite2/int8/latest/efficientdet_lite2.tflite int8 7.52 MB (7,515,971 B) b3f50554cb0ea559e90328845f7d9ba4d13c8bff372914d24e06bc8bb72fa896
efficientdet_lite2_float16.tflite object_detector/efficientdet_lite2/float16/latest/efficientdet_lite2.tflite float16 12.14 MB (12,138,859 B) 5d4ebec1029bc9907aeadb9e7b4ac9cb1da6a19d01ad375210a9ae18ba173302
efficientdet_lite2_float32.tflite object_detector/efficientdet_lite2/float32/latest/efficientdet_lite2.tflite float32 23.10 MB (23,096,891 B) ad2abbf2b4e10585e15176fd7b5ef03c28dda959ae26fc142549fdd1814db91d

Companion Files in Repo Root

File Role Size SHA-256
labelmap.txt Label map 0.7 KB (661 B) f8803ef7900160c629d570848dfda4175e21667bf7b71f73f8ece4938c9f2bf2

Download

pip install -U huggingface_hub
hf download litert-community/MediaPipe-ObjectDetector-EfficientDet-Lite2 efficientdet_lite2_int8.tflite --local-dir .
hf download litert-community/MediaPipe-ObjectDetector-EfficientDet-Lite2 efficientdet_lite2_float16.tflite --local-dir .
hf download litert-community/MediaPipe-ObjectDetector-EfficientDet-Lite2 efficientdet_lite2_float32.tflite --local-dir .

Inputs, Outputs, and Model Specifications

efficientdet_lite2_int8.tflite

  • Embedded TFLITE_METADATA: name ObjectDetector · version 1 · author MediaPipe · packed associated files: labels.txt
Direction Tensor Shape Type Notes
Input serving_default_images:0 [1, 448, 448, 3] uint8 image — Input image to be processed. — Normalization: mean=[127.5], std=[127.5] — Quantization: scale=0.0078125, zero_point=127
Output StatefulPartitionedCall:1 [1, 37629, 90] float32 score — The scores of the detected boxes. — Labels: labels.txt
Output StatefulPartitionedCall:0 [1, 37629, 4] float32 location — The locations of the detected boxes.

efficientdet_lite2_float16.tflite

  • Embedded TFLITE_METADATA: name ObjectDetector · version 1 · author MediaPipe · packed associated files: labels.txt
Direction Tensor Shape Type Notes
Input serving_default_images:0 [1, 448, 448, 3] float32 image — Input image to be processed. — Normalization: mean=[127.5], std=[127.5]
Output StatefulPartitionedCall:1 [1, 37629, 90] float32 score — The scores of the detected boxes. — Labels: labels.txt
Output StatefulPartitionedCall:0 [1, 37629, 4] float32 location — The locations of the detected boxes.

efficientdet_lite2_float32.tflite

  • Embedded TFLITE_METADATA: name ObjectDetector · version 1 · author MediaPipe · packed associated files: labels.txt
Direction Tensor Shape Type Notes
Input serving_default_images:0 [1, 448, 448, 3] float32 image — Input image to be processed. — Normalization: mean=[127.5], std=[127.5]
Output StatefulPartitionedCall:1 [1, 37629, 90] float32 score — The scores of the detected boxes. — Labels: labels.txt
Output StatefulPartitionedCall:0 [1, 37629, 4] float32 location — The locations of the detected boxes.

Use with MediaPipe Tasks

Python Quickstart

pip install -U mediapipe huggingface_hub
import mediapipe as mp
from huggingface_hub import hf_hub_download

model_path = hf_hub_download(
    repo_id="litert-community/MediaPipe-ObjectDetector-EfficientDet-Lite2",
    filename="efficientdet_lite2_int8.tflite",
)

BaseOptions = mp.tasks.BaseOptions
ObjectDetector = mp.tasks.vision.ObjectDetector
ObjectDetectorOptions = mp.tasks.vision.ObjectDetectorOptions
VisionRunningMode = mp.tasks.vision.RunningMode

options = ObjectDetectorOptions(
    base_options=BaseOptions(model_asset_path=model_path),
    running_mode=VisionRunningMode.IMAGE,
    score_threshold=0.5,
    max_results=5,
)

with ObjectDetector.create_from_options(options) as runner:
    mp_image = mp.Image.create_from_file("input.jpg")
    result = runner.detect(mp_image)
    for det in result.detections:
        cat = det.categories[0]
        print(f"{cat.category_name}: {cat.score:.3f} box={det.bounding_box}")

Cross-Platform Guides & Samples

Latency Benchmarks

End-to-end MediaPipe Task pipeline average latency reported by the Object Detector documentation:

Model Variant Device CPU Latency GPU Latency
EfficientDet-Lite2 (int8) Pixel 6 70.91 ms —
EfficientDet-Lite2 (float 16) Pixel 6 198.77 ms 47.31 ms
EfficientDet-Lite2 (float 32) Pixel 6 197.98 ms 41.15 ms

Intended Use, Limitations, and Fairness

  • Intended applications: General-purpose multi-object localization and classification across common COCO categories (people, vehicles, animals, household items, food, etc.) in still images, video, and camera streams.
  • Limitations & considerations:
    • Closed-vocabulary COCO detector (80 active object classes); objects outside COCO cannot be named.
    • Small objects, heavy occlusion, motion blur, or non-natural camera angles reduce detection recall.
    • Not intended for safety-critical perception (autonomous driving, medical imaging, surveillance) without domain-specific fine-tuning and evaluation.
    • Google does not publish a separate standalone fairness PDF for this model on the MediaPipe documentation page.

Provenance

Every model binary in this repository is byte-identical (unmodified FlatBuffer weights and embedded TFLITE_METADATA bytes) to the official artifact served by Google from storage.googleapis.com/mediapipe-models.

Repo File Upstream URL GCS Generation Upstream Last-Modified Pinned Bucket Version
efficientdet_lite2_int8.tflite efficientdet_lite2.tflite 1682636020984971 Thu, 27 Apr 2023 22:53:41 GMT /1/
efficientdet_lite2_float16.tflite efficientdet_lite2.tflite 1682636018681342 Thu, 27 Apr 2023 22:53:38 GMT /1/
efficientdet_lite2_float32.tflite efficientdet_lite2.tflite 1682636017570268 Thu, 27 Apr 2023 22:53:37 GMT /1/

License and Citation

  • License: apache-2.0 — Distributed under Apache License, Version 2.0 alongside the MediaPipe release artifacts.

Citation

@inproceedings{tan2020efficientdet,
  title     = {EfficientDet: Scalable and Efficient Object Detection},
  author    = {Tan, Mingxing and Pang, Ruoming and Le, Quoc V.},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2020}
}
Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Collection including litert-community/MediaPipe-ObjectDetector-EfficientDet-Lite2