ultralytics-pt-yolov3-vitis-ai-edge

This demo is only used for inference testing of Vitis AI v1.4 and quantitative compilation of DPU. It is compatible with the training results of ultralytis yolov3 v9.5.0 (it needs to use the model saving method of Pytorch V1.4)

envirment

• Yocto sdk 2020.2
• Vitis-AI V1.4 Docker-GPU

Attention

Training code using ultralytis yolov3 BUT the output part of the model weights in the code NEED SOME MODIFICATION.

weights saving part

# Save model
if (not opt.nosave) or (final_epoch and not opt.evolve):  # if save
    ckpt = {'epoch': epoch,
            'best_fitness': best_fitness,
            'training_results': results_file.read_text(),
            'model': deepcopy(de_parallel(model)).half(),
            'ema': deepcopy(ema.ema).half(),
            'updates': ema.updates,
            'optimizer': optimizer.state_dict(),
            'wandb_id': wandb_logger.wandb_run.id if wandb_logger.wandb else None}

    # Save last, best and delete
    torch.save(ckpt, last)
    torch.save(model, last_quant, _use_new_zipfile_serialization=False) # For quantization (用于量化的兼容版本)

    if best_fitness == fi:
        torch.save(ckpt, best)
        torch.save(model, best_quant, _use_new_zipfile_serialization=False) # For quantization (用于量化的兼容版本)
    if wandb_logger.wandb:
        if ((epoch + 1) % opt.save_period == 0 and not final_epoch) and opt.save_period != -1:
            wandb_logger.log_model(
                last.parent, opt, epoch, fi, best_model=best_fitness == fi)
    del ckpt