|
--- |
|
tags: |
|
- computer_vision |
|
- vision_models_playground |
|
- custom-implementation |
|
--- |
|
# **Vision Models Playground** |
|
This is a trained model from the Vision Models Playground repository. |
|
Link to the repository: https://github.com/Akrielz/vision_models_playground |
|
|
|
## **Model** |
|
This model is a custom implementation of **ResNetYoloV1** from the ```vision_models_playground.models.segmentation.yolo_v1``` module. |
|
Please look in the config file for more information about the model architecture. |
|
|
|
## **Usage** |
|
To load the torch model, you can use the following code snippet: |
|
|
|
```python |
|
import torch |
|
from vision_models_playground.utility.hub import load_vmp_model_from_hub |
|
|
|
|
|
model = load_vmp_model_from_hub("Akriel/ResNetYoloV1") |
|
|
|
x = torch.randn(...) |
|
y = model(x) # y will be of type torch.Tensor |
|
``` |
|
|
|
To load the pipeline that includes the model, you can use the following code snippet: |
|
|
|
```python |
|
from vision_models_playground.utility.hub import load_vmp_pipeline_from_hub |
|
|
|
pipeline = load_vmp_pipeline_from_hub("Akriel/ResNetYoloV1") |
|
|
|
x = raw_data # raw_data will be of type pipeline.input_type |
|
y = pipeline(x) # y will be of type pipeline.output_type |
|
``` |
|
|
|
## **Metrics** |
|
|
|
The model was evaluated on the following dataset: **YoloPascalVocDataset** from ```vision_models_playground.datasets.yolo_pascal_voc_dataset``` |
|
|
|
These are the results of the evaluation: |
|
- MulticlassAccuracy: 0.7241 |
|
- MulticlassAveragePrecision: 0.7643 |
|
- MulticlassAUROC: 0.9684 |
|
- Dice: 0.7241 |
|
- MulticlassF1Score: 0.7241 |
|
- LossTracker: 4.1958 |
|
|
|
|
|
## **Additional Information** |
|
The train and evaluation runs are also saved using tensorboard. You can use the following command to visualize the runs: |
|
|
|
```bash |
|
tensorboard --logdir ./model |
|
``` |
|
|
|
```bash |
|
tensorboard --logdir ./eval |
|
``` |