Multi-tasking model for grasping and classification (in progress)
you can see the model architecture in multi_task_models/grcn_multi_alex.py
Run this from inside the matlab_files directory
gdown --folder https://drive.google.com/drive/folders/1FBsL4MVpCpqCuobVhH1_23RLRHDfx0Y7?ths=true
python3 rsa.py [suffix of output file]
| Task | Recogniton | Grasping |
|---|---|---|
| Train Accuracy (%) | 99.02 | 83.65 |
| Test Accuracy (%) | 85.0 | 81.5 |
| Learning Rate | 301 | 283 |
| Epoch | 150 | 150 |
Size of divergent heads: 4 layers
Weighted Loss Ratio (Grasp : Classification): 1.5 : 0.5
Epochs: 150
Batch Size: 5
Grasp Accuracies - Training: 83.65 - Test: 81.5
Classification Accuracies - Training: 99.02 - Test: 85.0
Size of divergent heads: 4 layers
Weighted Loss Ratio (Grasp : Classification): 0.5 : 1.5
Epochs: 150
Batch Size: 5
Grasp Accuracies - Training: 77.9 - Test: 75.5
Classification Accuracies - Training: 97.98 - Test: 84.5
Size of divergent heads: 4 layers
Epochs: 130
Batch Size: 5
Grasp Accuracies - Training: 79.95 - Test: 79.5
Classification Accuracies - Training: 98.17 - Test: 82.75
Size of divergent heads: 1 layer
Epochs: 150
Batch Size: 2
Grasp Accuracies - Training: 72.4 - Test: 67.0
Classification Accuracies - Training: 98.53 - Test: 89.25
Size of divergent heads: 1 layer
Epochs: 150
Batch Size: 5
Grasp Accuracies - Training: 72.22 - Test: 75.75
Classification Accuracies - Training: 98.5 - Test: 82.75