canornot
canornot
感谢作者的贡献！ECBSR模型在DIV2K的bicubic的数据上PSNR能确实到达很高。为了更好的落地作者的算法，我们尝试用同样的模型训练一批仿真的DIV2K多帧Y数据（退化方案不仅是Bicubic，而是复杂多样的，并通过亚像素位移融入多帧信息），得出来的PSNR结果却降到26左右且很难上升到30以上，Loss也很难像bicubic一样降到3左右： ##===========Epoch: 8=============## Epoch:008, 0003200/3933696, loss: 7.0499, time: 13.528, psnr: 26.5676, ssim: 0.8217 Epoch:008, 0006400/3933696, loss: 6.9462, time: 8.626, psnr: 26.0487, ssim: 0.8200 Epoch:008, 0009600/3933696, loss: 6.9424, time: 8.789, psnr:...
第二阶段只微调线性层，还是类似LLaVa微调线性层加Vicuna啊？
In second stage finetuing, "we finetune our pretrained model with the curated high-quality image-textpairs". Does it mean only the linear projection layer is being finetuned in 2nd stage, similarly to...
For large networks, negative MAdds appear. It's because the caculated macs exceed the data range at line 53: param.nelement() returns python int64 but np.prod(info["out"][2:])) returns numpy np.int32 multiplication of these...
Original code: elif output_format == 'AL': output = f"Answer: The answer is {answer}. BECAUSE: {solution}" elif output_format == 'AE': output = f"Answer: The answer is {answer}. BECAUSE: {lecture}" Shall be...
Qwen2.5-VL-3B/7B/72B-Instruct模型针对图片输入，我用中文问可以用中文答，用英文问可以用英文答；但是对视频输入，我用中文问，并Prompt提示"请用中文回答“，但是模型还是只能用英文回答？？ 有哪位亲对视频输入成功测出来中文回答的吗？