wesfggfd/BirdCLEF-2023

For this competition, you'll use your machine-learning skills to identify Eastern African bird species by sound. Specifically, you'll develop computational solutions to process continuous audio data and recognize the species by their calls. The best entries will be able to train reliable classifiers with limited training data # Performance Improvements ```bash Baseline → Supervised Contrastive Learning: Absolute gain: 0.61600 − 0.61058 = 0.00542 Relative improvement: (0.00542 / 0.61058) × 100% ≈ 0.89% Supervised Contrastive Learning → NFNet-L0 Ensemble: Absolute gain: 0.75688 − 0.61600 = 0.14088 Relative improvement: (0.14088 / 0.61600) × 100% ≈ 22.9% Baseline → NFNet-L0 Ensemble (Overall): Absolute gain: 0.75688 − 0.61058 = 0.14630 Relative improvement: (0.14630 / 0.61058) × 100% ≈ 23.96% ``` **Dataset_V_1:huggingface.co **Pretrained_Model_V_1:huggingface.co - [kaggle] - [└── 📁 input/] - ├── 📁 birdclef-2023-v1/ - [│ ├── 📁 train_audio/] - [│ └── 📄 train_metadata.csv] - [│ └── 📁 test_soundscapes/] - [│] - ├── 📁 pretrained_models/ - [│ └── 📄 v1_bird_vocalization_classifier/] - [│ └── 📁 assets/] - [│ └── 📁 variables/] - [├── 📁 kaggle/] - [└── 📁 input/] - [│ └── BirdCLEF 2023] - [├── 📁 DATASETS] - │ └── bc2023-train-val-df - [│ ├── train/] - [│ ├── val/] - [├── 📁 NOTEBOOKS] - │ ├── BirdCLEF23 Supervised Contrastive Loss Training.ipynb - [theoretical explanation] (github.com) - [│ └── Split & Creating MelSpecs [Stage 1].ipynb] (kaggle.com) - [kaggle] - [└── 📁 input/] - ├── 📁 birdclef-2023/ - [├── 📁 DATASETS] - │ └── BirdCLEF2023-4th-models - theoretical explanation **V1 Best Score:0.61058** **V2 Best Score:0.61600** **V3 Best Score:0.75688** # BirdCLEF 2023 Competition Submission Code Analysis This document provides a detailed breakdown of three code versions (V1, V2, V3) submitted for the BirdCLEF 2023 competition, explaining their technical approaches, improvements, and performance outcomes. --- ## Version 1 (Score: 0.61058) ### Approach - **Framework**: TensorF …