Improve some text

src/traditional/features/README.md

@@ -241,27 +241,55 @@ The raw image files are not committed to Git.
 
 ## Validation
 
-| Feature | Shape | Data type |
-| --- | --- | --- |
+| Feature | Shape | Data type | Output length |
+| --- | --- | --- | --- |
 | HOG | `(6084,)` | `float32` | Fixed |
 | RGB colour histogram | `(96,)` | `float32` | Fixed |
 | Combined HOG and colour | `(6180,)` | `float32` | Fixed |
 | SIFT descriptors | `(N, 128)` | `float32` | Variable |
 
-Both methods were tested on images from different classes and produced fixed-length feature vectors.
+The HOG, colour, and combined features produced fixed-length vectors. SIFT produced a variable number of 128-dimensional descriptors.
 
-## Current Branch
+## Downloading Traditional Feature Files
 
-Current branch:
+The extracted traditional feature files are stored separately on OneDrive because the `outputs` folder is not tracked by Git.
 
-`Chaohao_TraditionalFeature3`
+Download the feature files from:
 
-This branch continues the pipeline by combining the HOG and colour features.
+[Traditional feature files on OneDrive](https://unsw-my.sharepoint.com/:f:/g/personal/z5528581_ad_unsw_edu_au/IgAsAxZV1fJQSbL1AAmSgGoaAU-ZRhVTJmVLsHFxH7leipA)
+
+After downloading, place the files in:
+
+```text
+outputs/traditional_features/
+```
+
+The expected files are:
+
+```text
+train_combined_features.npz
+validation_combined_features.npz
+test_combined_features.npz
+```
+
+Each `.npz` file contains:
+
+```text
+features
+labels
+file_paths
+```
+
+These files are generated by `features.ipynb` and can be loaded by `cache.ipynb` or `traditional_classifier.ipynb`.
+
+The expected folder structure is:
 
 ```text
-HOG:      6084
-Colour:     96
-Combined:    6180
+outputs/
+└── traditional_features/
+    ├── train_combined_features.npz
+    ├── validation_combined_features.npz
+    └── test_combined_features.npz
 ```
 
 ## Maintenance
@@ -270,6 +298,6 @@ When adding a new feature:
 
 - use the shared settings from `config.py`
 - accept a PIL image as input
-- return a one-dimensional `float32` NumPy array
-- keep the output length fixed
+- return `float32` NumPy arrays
+- clearly state whether the output is fixed-length or variable-length
 - test the method on images from different classes

src/traditional/features/color.ipynb

@@ -1,23 +1,13 @@
 {
  "cells": [
-  {
-   "cell_type": "markdown",
-   "id": "2698ae8b",
-   "metadata": {},
-   "source": [
-    "Extract an RGB colour histogram from each image.\n",
-    "\n",
-    "The red, green, and blue channel histograms are normalised and combined into one fixed-length feature vector."
-   ]
-  },
   {
    "cell_type": "markdown",
    "id": "7b1019cf",
    "metadata": {},
    "source": [
     "# Colour Histogram Feature Extraction\n",
     "\n",
-    "It extracts colour histogram features from the iNaturalist images."
+    "Extract colour histogram features from the iNaturalist images."
    ]
   },
   {

src/traditional/features/combine.ipynb

@@ -17,7 +17,7 @@
    "id": "d79d4428",
    "metadata": {},
    "source": [
-    "## 1.Import libraries"
+    "## 1. Import libraries"
    ]
   },
   {
@@ -44,9 +44,9 @@
    "id": "fb4d7786",
    "metadata": {},
    "source": [
-    "## 2.Load the shared project configuration\n",
+    "## 2. Load the shared project configuration\n",
     "\n",
-    "It finds the project root and imports `config.py`.\n",
+    "Find the project root and imports `config.py`.\n",
     "\n",
     "Using the shared configuration avoids personal absolute paths and keeps the notebook working on different computers."
    ]
@@ -94,9 +94,9 @@
    "id": "ed08813c",
    "metadata": {},
    "source": [
-    "## 3.Load and check a sample image\n",
+    "## 3. Load and check a sample image\n",
     "\n",
-    "It selects a sample image from the training set, checks that the file exists, loads it as an RGB image, and displays its basic information.\n",
+    "Select a sample image from the training set, checks that the file exists, loads it as an RGB image, and displays its basic information.\n",
     "\n",
     "Change `sample_index` to test another image."
    ]
@@ -188,11 +188,11 @@
    "id": "5ccbc08f",
    "metadata": {},
    "source": [
-    "## 4.Define the HOG feature function\n",
+    "## 4. Define the HOG feature function\n",
     "\n",
-    "This function uses the same HOG settings as `hog.ipynb`.\n",
+    "Use the same HOG settings as `hog.ipynb`.\n",
     "\n",
-    "It resizes the image, converts it to grayscale, and returns a fixed-length `float32` feature vector."
+    "Resize the image, converts it to grayscale, and returns a fixed-length `float32` feature vector."
    ]
   },
   {
@@ -224,11 +224,11 @@
    "id": "103c08d1",
    "metadata": {},
    "source": [
-    "## 5.Define the colour histogram function\n",
+    "## 5. Define the colour histogram function\n",
     "\n",
-    "This function uses the same RGB histogram method as `color.ipynb`.\n",
+    "Use the same RGB histogram method as `color.ipynb`.\n",
     "\n",
-    "It resizes the image, calculates a 32-bin histogram for each RGB channel, normalises each histogram, and combines them into one `float32` feature vector."
+    "Resize the image, calculates a 32-bin histogram for each RGB channel, normalises each histogram, and combines them into one `float32` feature vector."
    ]
   },
   {
@@ -268,9 +268,9 @@
    "id": "7ab0b3dd",
    "metadata": {},
    "source": [
-    "## 6.Define the combined feature function\n",
+    "## 6. Define the combined feature function\n",
     "\n",
-    "This function extracts the HOG feature and RGB colour histogram from the same image.\n",
+    "Extract the HOG feature and RGB colour histogram from the same image.\n",
     "\n",
     "The two feature vectors are then joined into one fixed-length `float32` vector."
    ]
@@ -298,11 +298,9 @@
    "id": "a52de3be",
    "metadata": {},
    "source": [
-    "## 7.Test the combined feature\n",
+    "## 7. Test the combined feature\n",
     "\n",
-    "It extracts the HOG, colour, and combined features from the sample image.\n",
-    "\n",
-    "It checks the output shape, data type, and whether the combined feature contains invalid values."
+    "Extract the three feature vectors from one sample image and check the final output."
    ]
   },
   {
@@ -342,11 +340,9 @@
    "id": "980f7173",
    "metadata": {},
    "source": [
-    "## 8.Check the combined feature sections\n",
-    "\n",
-    "It checks that the combined feature keeps the correct order.\n",
+    "## 8. Check the feature order\n",
     "\n",
-    "The first part should match the HOG feature, and the remaining part should match the colour feature."
+    "Check that the HOG values come first and the colour values come after them."
    ]
   },
   {
@@ -391,7 +387,7 @@
    "id": "951a4f93",
    "metadata": {},
    "source": [
-    "## 9.Test images from different classes\n",
+    "## 9. Test images from different classes\n",
     "\n",
     "It selects sample images from different labels and checks whether the combined feature has the same shape and data type for every image."
    ]
@@ -458,7 +454,7 @@
    "id": "96d2157d",
    "metadata": {},
    "source": [
-    "## 10.Final validation\n",
+    "## 10. Final validation\n",
     "\n",
     "It performs a final check of the combined feature before it is used in the feature caching pipeline."
    ]

src/traditional/features/fearures.ipynb → src/traditional/features/features.ipynb

@@ -1558,9 +1558,9 @@
    "id": "92d4e928",
    "metadata": {},
    "source": [
-    "## 21. Final validation\n",
+    "## 21. Final check\n",
     "\n",
-    "Perform the final checks on all extracted feature matrices and saved output files."
+    "Check the final feature files before using them for classifier training."
    ]
   },
   {

src/traditional/features/sift.ipynb

@@ -94,7 +94,7 @@
    "source": [
     "## 3. Load and check a sample image\n",
     "\n",
-    "It selects one image from the training set, checks the image path, and displays its basic information."
+    "Select one image from the training set, check the image path, and display its basic information."
    ]
   },
   {
@@ -186,7 +186,7 @@
    "source": [
     "## 4. Define the SIFT feature function\n",
     "\n",
-    "It resizes the image, converts it to grayscale, and extracts SIFT keypoints and descriptors.\n",
+    "Resizes the image, converts it to grayscale, and extracts SIFT keypoints and descriptors.\n",
     "\n",
     "`max_features=500` requests approximately 500 of the strongest keypoints.\n",
     "\n",
@@ -251,7 +251,7 @@
    "source": [
     "## 5. Test the SIFT feature\n",
     "\n",
-    "This cell extracts SIFT keypoints and descriptors from the sample image and checks the output."
+    "Extract SIFT keypoints and descriptors from the sample image and check the output."
    ]
   },
   {
@@ -292,7 +292,7 @@
    "source": [
     "## 6. Visualise the SIFT keypoints\n",
     "\n",
-    "It draws the detected SIFT keypoints on the resized sample image."
+    "Draw the detected SIFT keypoints on the resized sample image."
    ]
   },
   {
@@ -342,7 +342,7 @@
    "source": [
     "## 7. Check the descriptor values\n",
     "\n",
-    "It checks the value range and basic statistics of the SIFT descriptors."
+    "Check the value range and basic statistics of the SIFT descriptors."
    ]
   },
   {
@@ -381,7 +381,7 @@
    "source": [
     "## 8. Visualise one SIFT descriptor\n",
     "\n",
-    "It displays the 128 values of the first SIFT descriptor.\n",
+    "Display the 128 values of the first SIFT descriptor.\n",
     "\n",
     "Each descriptor describes the local gradient information around one detected keypoint."
    ]
@@ -423,7 +423,7 @@
    "source": [
     "## 9. Test images from different classes\n",
     "\n",
-    "This cell tests the SIFT feature extractor on images from different labels.\n",
+    "Test the SIFT feature extractor on images from different labels.\n",
     "\n",
     "The number of descriptors may change between images, but every descriptor must contain 128 values."
    ]
@@ -513,7 +513,7 @@
    "source": [
     "## 10. Display the test results\n",
     "\n",
-    "This cell places the SIFT test results into a table for easier comparison."
+    "It places the SIFT test results into a table for easier comparison."
    ]
   },
   {
@@ -628,7 +628,7 @@
    "id": "e587cb77",
    "metadata": {},
    "source": [
-    "##  11.Visualise keypoints from different classes\n",
+    "##  11. Visualise keypoints from different classes\n",
     "\n",
     "Display the detected SIFT keypoints for the selected images from different classes."
    ]
@@ -746,7 +746,7 @@
    "source": [
     "## 12. Check the descriptor dimensions\n",
     "\n",
-    "It confirms that all tested SIFT descriptors contain 128 values."
+    "Confirm that all tested SIFT descriptors contain 128 values."
    ]
   },
   {
@@ -793,7 +793,7 @@
     "\n",
     "SIFT detects a different number of keypoints in different images.\n",
     "\n",
-    "It compares the descriptor row counts produced by the tested images."
+    "Compare the descriptor row counts produced by the tested images."
    ]
   },
   {
@@ -832,7 +832,7 @@
    "source": [
     "## 14. Final validation\n",
     "\n",
-    "It performs the final checks for the sample SIFT output.\n",
+    "Perform the final checks for the sample SIFT output.\n",
     "\n",
     "The number of descriptor rows may vary between images, but every descriptor must contain 128 values."
    ]
@@ -891,17 +891,11 @@
    "source": [
     "## Summary\n",
     "\n",
-    "SIFT keypoints and descriptors were successfully extracted from the sample images.\n",
+    "SIFT was used to detect local keypoints and extract 128-dimensional descriptors.\n",
     "\n",
-    "- Each SIFT descriptor contains 128 values.\n",
-    "- The descriptor data type is `float32`.\n",
-    "- The number of keypoints can be different for each image.\n",
-    "- `max_features=500` requests approximately 500 of the strongest keypoints.\n",
-    "- The returned number may be slightly above 500 because one location can produce keypoints with multiple orientations.\n",
-    "- The keypoint count matches the number of descriptor rows.\n",
-    "- No missing or infinite values were found.\n",
+    "Different images produced different numbers of descriptors, so the raw SIFT output does not have a fixed length. Because of this, it cannot be directly added to the current HOG and colour feature vector.\n",
     "\n",
-    "Unlike HOG and the RGB colour histogram, SIFT does not directly produce one fixed-length feature vector for each image. A method such as Bag of Visual Words can later convert the SIFT descriptors into a fixed-length representation for classifier training."
+    "A method such as Bag of Visual Words would be needed before using SIFT with the traditional classifiers."
    ]
   }
  ],