License clarification for PP-PicoDet pretrained and fine-tuned weight redistribution in a commercial iOS app

We are evaluating the official picodet_s_416_coco_lcnet.pdparams checkpoint published by PaddleDetection. We understand that the PaddleDetection repository is released under Apache-2.0 and that PP-PicoDet is a PaddleDetection-developed model intended for mobile and industrial deployment.

Could a PaddlePaddle maintainer please confirm in writing whether the following uses are permitted?

1. Commercial use of the official PP-PicoDet-S 416 .pdparams checkpoint.
2. Fine-tuning that checkpoint on our own authorized basketball videos.
3. Converting the fine-tuned derivative weights to ONNX and Core ML.
4. Embedding and redistributing the resulting derivative Core ML model inside a closed-source, paid or subscription-based iOS App Store application.
5. Retaining the derivative model on users' devices for fully on-device inference.

If these uses are permitted, does Apache-2.0 apply to the official checkpoint and resulting derivative weights, and what copyright, attribution, NOTICE, model-card, or training-data notices must be included in the app or accompanying documentation?

The exact official checkpoint URL is: https://paddledet.bj.bcebos.com/models/picodet_s_416_coco_lcnet.pdparams

The checkpoint SHA-256 is: f5d0de89f76a17169badaadff35304eb9f514d5e0c54c44b566d94bba0ea96d2

We would also appreciate confirmation of the checkpoint's training-data lineage and whether PaddlePaddle imposes any additional terms arising from that training data.