NVIDIA TAO models are no longer available. Your existing models will keep working (incl. deployment), but you can no longer retrain this model. You can switch to another model architecture via Choose a different model, or see the docs to learn more.
Object detection settings
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
Number of training cycles should be a number
Minimum learning rate should be a number
Maximum learning rate should be a number
Hue should be a number
Saturation should be a number
Exposure should be a number
Vertical flip should be a number
Horizontal flip should be a number
Jitter should be a number
Randomize input shape period should be a number
Please provide a valid number for the train/validate split (between 0 and 1)
%
Soft start should be a number
Annealing should be a number
Cluster IoU threshold should be a number
Confidence threshold should be a number
Batch size should be a number
Matching neutral box IoU should be a number
Localization loss should be a number
Negative objectiveness loss should be a number
Classification loss should be a number
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Freeze blocks should not be empty
Fine tune training cycles should be a number
Fine tune minimum learning rate should be a number
Fine tune maximum learning rate should be a number
Fine tune soft start should be a number
Fine tune annealing should be a number
Neural network architecture
Input layer (150,528 features)
Choose a different model
Output layer (2 classes)
Model
Model version:
Last training performance (validation set)
Precision score
58.4%
Confusion matrix (validation set)
Loading...
Metric | Value |
---|---|
mAP | 0.10 |
mAP@[IoU=50] | 0.39 |
mAP@[IoU=75] | 0.02 |
mAP@[area=small] | 0.00 |
mAP@[area=medium] | 0.05 |
mAP@[area=large] | 0.23 |
Recall@[max_detections=1] | 0.17 |
Recall@[max_detections=10] | 0.18 |
Recall@[max_detections=100] | 0.18 |
Recall@[area=small] | 0.02 |
Recall@[area=medium] | 0.12 |
Recall@[area=large] | 0.30 |
On-device performance
Engine:
Inferencing time
N/A
Peak RAM usage
2.5M
Flash usage
1.2M
This model won't run on MCUs.
Last training performance (validation set)
Precision score
54.6%
Confusion matrix (validation set)
Loading...
Metric | Value |
---|---|
mAP | 0.11 |
mAP@[IoU=50] | 0.39 |
mAP@[IoU=75] | 0.01 |
mAP@[area=small] | 0.10 |
mAP@[area=medium] | 0.06 |
mAP@[area=large] | 0.18 |
Recall@[max_detections=1] | 0.17 |
Recall@[max_detections=10] | 0.22 |
Recall@[max_detections=100] | 0.22 |
Recall@[area=small] | 0.20 |
Recall@[area=medium] | 0.16 |
Recall@[area=large] | 0.31 |
On-device performance
Engine:
Inferencing time
6268 ms.
Peak RAM usage
0.0K
Flash usage
3.9M
This model won't run on MCUs.
Calculated arena size is >6MB