Tree Counting and Species Identification

### 📦 Dataset Description #### 🟢 Dataset 1: **Homogeneous Tree Species Dataset** This dataset consists of images from **four different tree species**, organized into three subsets: * **Training*** **Validation*** **Test** The number of samples per class has been balanced across all subsets using **data augmentation techniques**. This balancing ensures that the model does not become biased toward any specific tree class and maintains **equal sensitivity across all categories**. --- #### 🔴 Dataset 2: **Tree Centers from TreeCountNet** This dataset contains the predicted **coordinates of tree centers** generated by a **regression-based deep network called TreeCountNet**. The training process for TreeCountNet included: 1. Initial training on **incomplete dot-map annotations**2. Further improvement using **pseudo-labeling techniques** This network achieved **over 90% accuracy** in locating tree positions. The predicted coordinates are stored in the `dots_csv` folder and are later used for generating red points in the test images. --- #### 🔵 Dataset 3: **Images with Red Dots at Tree Centers** This dataset contains modified versions of the training and validation images from **Dataset 1**, where: * A **red dot with a radius of 5 pixels** is drawn at the **center of each labeled bounding box**.* These red points visually represent the exact location of each labeled tree.* For the **test subset**, red dots are drawn using the coordinates provided by the TreeCountNet predictions (i.e., from **Dataset 2**) instead of manual labels. --- ### 🧾 CSV Files Structure For all three subsets (train, validation, and test), the corresponding CSV annotation files follow this format: | filename | class | xmin | ymin | xmax | ymax || ---------------- | ----- | ----- | ----- | ----- | ----- || `image_name.tif` | 1–4 | float | float | float | float | These files are used for: * Training object detection models* Visualizing labeled data* Evaluating prediction accuracy ---GitHub Link : https://github.com/beheshtiniya/-Tree-Counting-and-Species-Classification-in-Dense-Forests-Using-Aerial-Imagery-and-Hybrid-Regressio