Edge Impulse Inc. / Tutorial: object detection Public
Target: Nvidia Jetson Orin Nano Clone this project

Neural Network settings

Switch to Keras (expert) mode Switch to visual (simple) mode Edit block locally

Training settings

Please provide a valid number of training cycles (numeric only)
Please provide a valid number for the learning rate (between 0 and 1)
Please provide a valid training processor option

Augmentation settings

Advanced training settings

Neural network architecture

Save
sys.path.append('./resources/libraries') import ei_tensorflow.object_detection EPOCHS = args.epochs or 25 LEARNING_RATE = args.learning_rate or 0.015 ei_tensorflow.object_detection.train(classes, LEARNING_RATE, EPOCHS, train_dataset, validation_dataset) # The 'model' variable is expected to be a Keras model. For this model type there is no keras model, # so we just set it to None. model = None
Input layer (307,200 features)
MobileNetV2 SSD FPN-Lite 320x320
Output layer (2 classes)

Model

Model version:

Last training performance (validation set)

Precision score
90.8%

Metrics (validation set)

Metric Value
mAP 0.62
mAP@[IoU=50] 1.00
mAP@[IoU=75] 0.68
mAP@[area=small] -1.00
mAP@[area=medium] 0.77
mAP@[area=large] 0.58
Recall@[max_detections=1] 0.70
Recall@[max_detections=10] 0.70
Recall@[max_detections=100] 0.70
Recall@[area=small] -1.00
Recall@[area=medium] 0.80
Recall@[area=large] 0.64

Data explorer (full training set)

Loading...

On-device performance

Engine:
EON™ Compiler TensorFlow Lite
Inferencing time
40 ms.
Flash usage
11.0M
This model won't run on MCUs. Calculated arena size is >6MB
Latency estimation for your chosen target is not currently supported
Help & documentation Get help on the forum Send feedback Contact sales
Join us
Terms of Service Privacy policy