Conversation
There was a problem hiding this comment.
Pull request overview
This PR adds a new Jupyter notebook titled "Dilshan2" that provides a comprehensive guide for training GlueFactory/LightGlue models on Google Colab. The notebook includes setup instructions, dataset preparation from Google Drive, training configuration, model evaluation, and inference examples.
- Adds a complete training workflow notebook for GlueFactory/LightGlue models
- Includes Google Colab-specific setup with Google Drive integration
- Provides end-to-end examples from installation through model inference
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| "id": "4fef2ad5", | ||
| "metadata": {}, | ||
| "source": [ | ||
| "### 5.3 Start Training\n", |
There was a problem hiding this comment.
Section numbering is inconsistent. This is labeled as "5.3" but should be "6.3" to match the parent section numbering.
| "### 5.3 Start Training\n", | |
| "### 6.3 Start Training\n", |
| "id": "395da816", | ||
| "metadata": {}, | ||
| "source": [ | ||
| "### 6.1 Setup Training Datasets in Google Drive\n", |
There was a problem hiding this comment.
The section is numbered "6. Training Setup" but the subsection is labeled "6.1 Setup Training Datasets in Google Drive". The header at line 707 indicates this is actually about setting up training datasets, which is inconsistent with the earlier section numbering. Consider renumbering this subsection or clarifying the hierarchy.
| "### 6.1 Setup Training Datasets in Google Drive\n", | |
| "### 6.1 Training Dataset Setup in Google Drive\n", |
| " # Create unique filename: sequence_name + original_filename\n", | ||
| " # Example: v_abstract_1.ppm instead of just 1.ppm\n", | ||
| " new_filename = f\"{seq}_{img_file.name}\"\n", | ||
| " destination = os.path.join(TRAIN_IMAGES_DIR, new_filename)\n", |
There was a problem hiding this comment.
The code comment states "Create unique filename: sequence_name + original_filename" followed by an example, but the actual code implementation uses string formatting. Consider adding validation to ensure the new filename doesn't already exist before copying, to prevent silent overwrites in edge cases.
| " destination = os.path.join(TRAIN_IMAGES_DIR, new_filename)\n", | |
| " destination = os.path.join(TRAIN_IMAGES_DIR, new_filename)\n", | |
| " if os.path.exists(destination):\n", | |
| " print(f\"⚠️ Warning: {destination} already exists. Skipping copy to avoid overwrite.\")\n", | |
| " continue\n", |
| "- **Dataset:** Your HPatches images (~5000 image pairs)\n", | ||
| "- **Method:** Synthetic homographies for data augmentation\n", | ||
| "- **Model:** LightGlue with SuperPoint features\n", | ||
| "- **Epochs:** 30 (adjustable)\n", |
There was a problem hiding this comment.
The documentation incorrectly states "30" epochs when the configuration sets it to 20. This should be corrected to reflect the actual value used in the notebook.
| "- **Epochs:** 30 (adjustable)\n", | |
| "- **Epochs:** 20 (adjustable)\n", |
| "id": "9ec47a7d", | ||
| "metadata": {}, | ||
| "source": [ | ||
| "### 5.2 Create Training Configuration File" |
There was a problem hiding this comment.
Section numbering is inconsistent. This is labeled as "5.2" but follows section "6" (Training Setup). The section should be numbered as "6.2" or the previous sections should be renumbered.
| "### 5.2 Create Training Configuration File" | |
| "### 6.2 Create Training Configuration File" |
No description provided.