convnextv2-basketball-round-classifier

ConvNeXtV2 Basketball Round-State Classifier

This model classifies a single basketball broadcast frame as either inside or outside an active round/play segment. It was locally fine-tuned from timm/convnextv2_base.fcmae_ft_in22k_in1k.

Model details

• Architecture: ConvNeXtV2 Base
• Task: binary image classification
• Resolution: 224 × 224 RGB
• Classes: 0 = not_in_round, 1 = in_round
• Best validation accuracy: 89.21%
• First epoch reaching the best score: epoch 17
• Base model license: CC BY-NC 4.0

The reported accuracy is from the local validation split and has not been independently reproduced on a public benchmark.

Intended use

Non-commercial research and prototyping for basketball broadcast segmentation and editing assistance. Commercial use is not permitted under the inherited CC BY-NC 4.0 license without obtaining any additional permission required from the relevant rights holders.

Training procedure

The original 1,000-class classification head was replaced with a two-class head. Training used staged backbone freezing/unfreezing, AdamW and cosine learning-rate scheduling.

• Batch size: 32
• Initial learning rate: 1e-4
• Weight decay: 1e-4
• Image size: 224
• Normalization mean: [0.485, 0.456, 0.406]
• Normalization standard deviation: [0.229, 0.224, 0.225]

Training-data disclosure

Training images are not included. They were locally extracted from basketball broadcast footage and grouped into not_in_round and in_round. Some footage originated from publicly accessible Bilibili videos. Public accessibility is not a grant of redistribution rights, so this repository does not contain the source footage, frames, audio, subtitles, uploader information or platform metadata.

The source material has not undergone complete work-by-work copyright clearance. Users are responsible for assessing compliance with copyright, privacy, publicity and platform rules. Rights holders may request review or removal through the repository discussion/contact channel.

Usage

pip install -r requirements.txt
python inference.py path/to/frame.jpg --checkpoint model.pth

The PyTorch checkpoint contains model_state_dict plus training metadata. Load pickle-based files only from trusted sources.

Limitations

• A single frame may not contain enough temporal information to determine play state.
• Broadcast overlays, replays and close-up shots may reduce accuracy.
• Generalization to unfamiliar leagues, production styles and camera systems is unverified.
• The model may learn broadcaster-specific visual patterns.
• Do not use it for surveillance or identity-related decisions.

License

This derivative is released under Creative Commons Attribution-NonCommercial 4.0 International, consistent with the base model's current Hugging Face license metadata. Attribution must identify the timm base model and original ConvNeXtV2 work. See LICENSE.md.