| --- |
| language: id |
| tags: |
| - image-classification |
| - resnet18 |
| - pytorch |
| - computer-vision |
| license: mit |
| datasets: |
| - sumn2u/garbage-classification-v2 |
| metrics: |
| - accuracy |
| --- |
| |
| # Garbage Classification |
|
|
| Model ini adalah versi **ResNet18** yang di-fine-tune menggunakan dataset **Garbage Classification v2** dari Kaggle. Model ini mencapai hasil berikut pada dataset evaluasi (25% split): |
|
|
| - **Loss:** 0.0020 |
| - **Accuracy:** 0.9364 |
|
|
| ## Model description |
| Model klasifikasi gambar untuk membedakan jenis-jenis sampah guna mendukung sistem pemilahan otomatis. Menggunakan arsitektur ResNet18 dengan teknik Transfer Learning. |
|
|
| ## Intended uses & limitations |
| Digunakan untuk mengidentifikasi material tunggal dalam gambar. Limitasi mencakup kesulitan pada latar belakang yang ramai atau objek yang bertumpuk. |
|
|
| ## Training and evaluation data |
| Dataset menggunakan 12,260 gambar yang dibagi menjadi 75% Training (9,195 gambar) dan 25% Testing (3,065 gambar). |
|
|
| ## Training procedure |
|
|
| ### Training hyperparameters |
| The following hyperparameters were used during training: |
| - **learning_rate:** 0.0001 |
| - **train_batch_size:** 32 |
| - **eval_batch_size:** 32 |
| - **optimizer:** Adam with weight_decay=1e-4 |
| - **lr_scheduler_type:** StepLR (step=5, gamma=0.5) |
| - **num_epochs:** 15 |
| |
| ### Training results |
| | Epoch | Training Loss | Train Accuracy | |
| |-------|---------------|----------------| |
| | 1 | 0.5585 | 82.53% | |
| | 3 | 0.0790 | 97.77% | |
| | 6 | 0.0121 | 99.85% | |
| | 9 | 0.0043 | 99.98% | |
| | 12 | 0.0029 | 99.99% | |
| | 15 | 0.0020 | 100.00% | |
| |
| ## Framework versions |
| - **Pytorch:** 2.10.0+cu128 |
| - **Torchvision:** 0.25.0+cu128 |
| - **Python:** 3.12.12 |
| |
| ## Author |
| **Ina Alyani** |
| |
