Skip to content

openvino-pedestrian-detection.md

Latest commit

 

History

History
236 lines (204 loc) · 4.85 KB

openvino-pedestrian-detection.md

File metadata and controls

236 lines (204 loc) · 4.85 KB
Raw

Hardware Accelerated pedestrian detection and attributes

OpenVino Pedestrian detection with attributes in ~50ms on an intel cpu!

Warning! You must be using a 6, 7, or 8th generation intel cpu with ubuntu see here for more info

  • This model leverages Intel's open vino hardware acceleration for extremely fast inference times. Make sure to read the warning above to ensure you get the most speed out of this!

Attributes demo!

Speed

Hardware Inference Time (Milliseconds)
CPU 35 (intel i-7) + 9ms for attributes

Accuracy

Detection Accuracy
Person 88% AP
Attributes ~99% (if person detected)

Detector

Model Description
Person tensorflow based SSD (mobilenet) converted to openvino format

  • trained on an internal dataset with 3000 pedestrians to detect (in outside and inside scenes with camera at eye level or higher)

  • internally, the container runs a version of tensorflow serving for use with openvino with an http server in front of it

Pedestrian attributes

  • Attributes include

    • 6 point color palette of pedestrian
    • top and bottom color of pedestrian (great for search!)
    • has_backpack
    • has_bag
    • has_coat_jacket
    • has_hat
    • has_longhair
    • has_longpants
    • has_longsleeves
    • is_male

  • people slices are size 80w x 160h

Example Attributes!

Example Detection!

Run

#cpu
docker run -ti \\
-p 9090:9090 \\
sugarkubes/openvino-pedestrian-detection:cpu

ENV Variables

Variable Default
PORT 8080
HOST 0.0.0.0
GRPC_PORT 9001
GRPC_HOST 0.0.0.0
BASIC_AUTH_USERNAME ""
BASIC_AUTH_PASSWORD ""

Routes

GET / GET /health GET /healthz

  • Responds with a 200 for healthcheck

POST /predict

  • Example:
curl -X POST \\
http://0.0.0.0:9090/predict \\
-H 'Content-Type: application/json' \\
  -d '{ "get_attributes": true, "url": "https://s3.us-west-1.wasabisys.com/public.sugarkubes/repos/sugar-cv/object-detection/friends.jpg" }'

Example Output

{
  "image_size": [
    1200,
    800
  ],
  "pedestrians": [
    {
      "bottom_color": [
        158,
        150,
        131
      ],
      "confidence": "1.0",
      "has_backpack": false,
      "has_bag": false,
      "has_coat_jacket": false,
      "has_hat": false,
      "has_longhair": false,
      "has_longpants": false,
      "has_longsleeves": true,
      "is_male": false,
      "palette": [
        [
          40,
          41,
          40
        ],
        [
          135,
          141,
          143
        ],
        [
          71,
          71,
          69
        ],
        [
          106,
          108,
          108
        ],
        [
          247,
          248,
          249
        ],
        [
          171,
          174,
          174
        ]
      ],
      "top_color": [
        52,
        55,
        53
      ],
      "x1": 1033,
      "x2": 1190,
      "y1": 126,
      "y2": 804
    },
    {
      "bottom_color": [
        106,
        84,
        96
      ],
      "confidence": "1.0",
      "has_backpack": false,
      "has_bag": false,
      "has_coat_jacket": false,
      "has_hat": false,
      "has_longhair": false,
      "has_longpants": false,
      "has_longsleeves": true,
      "is_male": false,
      "palette": [
        [
          87,
          87,
          89
        ],
        [
          113,
          116,
          119
        ],
        [
          37,
          38,
          38
        ],
        [
          62,
          60,
          62
        ],
        [
          142,
          151,
          161
        ],
        [
          190,
          199,
          208
        ]
      ],
      "top_color": [
        58,
        59,
        57
      ],
      "x1": 691,
      "x2": 848,
      "y1": 183,
      "y2": 777
    },
    ... others
  ],
  "inference_time": 45.057
}

ENV Variables

Variable Default
PORT 8080
HOST 0.0.0.0
GPU "" (true for GPU version)
GPU_FRACTION 0.25 (25% of the gpu will be allocated to this model)
BASIC_AUTH_USERNAME ""
BASIC_AUTH_PASSWORD ""

Validated Hosts

  • intel i-3, i-5, i-7

UI

  • UI is located at localhost:9090/tester/index.html