Bug: inference custom audio

[luosiwu]

推理目标音频的时候，结果会少一部分帧，是不是有bug呢

[kyusuncho]

Hi, could you provide some details of which frames you are missing?

[luosiwu]

iterations = process_until // batch_size if process_until % batch_size != 0: iterations += 1 total_time = 0

render image

for idx in tqdm(range(iterations), desc="Rendering progress",total = iterations):
    try:
        viewpoint_cams = next(loader)
        output = render_from_batch(viewpoint_cams, gaussians, pipeline, 
                                random_color= False, stage='fine',
                                batch_size=batch_size, visualize_attention=False, only_infer=True)
    except:
        break
    total_time += output["inference_time"]
    image.append(output["rendered_image_tensor"].cpu())
    gt.append(output["gt_tensor"].cpu())

我debug了下，看起来是batch_size的原因，batch_size=1就没有该问题，当然更好的方式是优化一下代码

[lokvke]

iterations = process_until // batch_size if process_until % batch_size != 0: iterations += 1 total_time = 0 #render image for idx in tqdm(range(iterations), desc="Rendering progress",total = iterations): try: viewpoint_cams = next(loader) output = render_from_batch(viewpoint_cams, gaussians, pipeline, random_color= False, stage='fine', batch_size=batch_size, visualize_attention=False, only_infer=True) except: break total_time += output["inference_time"] image.append(output["rendered_image_tensor"].cpu()) gt.append(output["gt_tensor"].cpu()) 我debug了下，看起来是batch_size的原因，batch_size=1就没有该问题，当然更好的方式是优化一下代码

请问推理的效果和速度怎么样