预训练模型运行不起来啊

[wuchaoAAA]

1.加载预训练模型： model_path=("./SwinJSCC/SwinJSCC w- SA/SwinJSCC_w_SA_AWGN_HRimage_snr_psnr_C96.model")
load_weights(model_path)

2.给定的命令：--trainset DIV2K --testset kodak --distortion-metric MSE --model SwinJSCC_w/_SA --channel-type awgn --C 96 --multiple-snr 1,4,7,10,13 --model_size base

3.报错缺少权重：Traceback (most recent call last):
File "F:\Paper2\swintranformer2_runbase\main.py", line 348, in
load_weights(model_path)
File "F:\Paper2\swintranformer2_runbase\main.py", line 160, in load_weights
net.load_state_dict(pretrained, strict=True)
File "D:\InstallSoftware\anaconda\envs\bpgan_env_3\lib\site-packages\torch\nn\modules\module.py", line 1482, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for SwinJSCC:
Missing key(s) in state_dict: "encoder.bm_list1.0.fc.0.weight", "encoder.bm_list1.0.fc.0.bias", "encoder.bm_list1.0.fc.2.weight", "encoder.bm_list1.0.fc.2.bias", "encoder.bm_list1.0.fc.4.weight", "encoder.bm_list1.0.fc.4.bias", "encoder.bm_list1.1.fc.0.weight", "encoder.bm_list1.1.fc.0.bias", "encoder.bm_list1.1.fc.2.weight", "encoder.bm_list1.1.fc.2.bias", "encoder.bm_list1.1.fc.4.weight", "encoder.bm_list1.1.fc.4.bias", "encoder.bm_list1.2.fc.0.weight", "encoder.bm_list1.2.fc.0.bias", "encoder.bm_list1.2.fc.2.weight", "encoder.bm_list1.2.fc.2.bias", "encoder.bm_list1.2.fc.4.weight", "encoder.bm_list1.2.fc.4.bias", "encoder.bm_list1.3.fc.0.weight", "encoder.bm_list1.3.fc.0.bias", "encoder.bm_list1.3.fc.2.weight", "encoder.bm_list1.3.fc.2.bias", "encoder.bm_list1.3.fc.4.weight", "encoder.bm_list1.3.fc.4.bias", "encoder.bm_list1.4.fc.0.weight", "encoder.bm_list1.4.fc.0.bias", "encoder.bm_list1.4.fc.2.weight", "encoder.bm_list1.4.fc.2.bias", "encoder.bm_list1.4.fc.4.weight", "encoder.bm_list1.4.fc.4.bias", "encoder.bm_list1.5.fc.0.weight", "encoder.bm_list1.5.fc.0.bias", "encoder.bm_list1.5.fc.2.weight", "encoder.bm_list1.5.fc.2.bias", "encoder.bm_list1.5.fc.4.weight", "encoder.bm_list1.5.fc.4.bias", "encoder.bm_list1.6.fc.0.weight", "encoder.bm_list1.6.fc.0.bias", "encoder.bm_list1.6.fc.2.weight", "encoder.bm_list1.6.fc.2.bias", "encoder.bm_list1.6.fc.4.weight", "encoder.bm_list1.6.fc.4.bias", "encoder.sm_list1.0.weight", "encoder.sm_list1.0.bias", "encoder.sm_list1.1.weight", "encoder.sm_list1.1.bias", "encoder.sm_list1.2.weight", "encoder.sm_list1.2.bias", "encoder.sm_list1.3.weight", "encoder.sm_list1.3.bias", "encoder.sm_list1.4.weight", "encoder.sm_list1.4.bias", "encoder.sm_list1.5.weight", "encoder.sm_list1.5.bias", "encoder.sm_list1.6.weight", "encoder.sm_list1.6.bias", "encoder.sm_list1.7.weight", "encoder.sm_list1.7.bias".

作者，麻烦问一下，为什么会报错呢，实在没找出问题所在，需要帮忙

[semcomm]

最早训练模型只考虑了snr自适应参数，bm_list1和sm_list1是rate adaptive参数，所以您加载的snr自适应模型没有这些参数。如果您只需要snr自适应，加载模型时忽略这部分参数即可。

[wuchaoAAA]

非常感谢您的帮忙，已经运行起来啦。但是我又遇见一个小问题
1.加载预训练模型：model_path=("./SwinJSCC/SwinJSCC w- SA&RA/SwinJSCC_w_SAandRA_AWGN_HRimage_cbr_psnr_snr.model")

2.测试命令：--trainset DIV2K --testset kodak --distortion-metric MSE --model SwinJSCC_w/_SAandRA --channel-type awgn --C 32,64,96,128,192 --multiple-snr 0,4,8 --model_size base

3.出现的结果是SNR: [[0.0, 0.0, 0.0, 0.0, 0.0], [4.0, 4.0, 4.0, 4.0, 4.0], [8.0, 8.0, 8.0, 8.0, 8.0]]
CBR: [[0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125], [0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125], [0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125]]
PSNR: [[20.404406030972797, 25.127057949701946, 25.701947053273518, 26.854474862416584, 29.68847068150838], [20.880515893300373, 26.36423667271932, 26.87330198287964, 28.887059609095257, 33.189711252848305], [21.143070936203003, 26.90866978963216, 27.359466234842937, 29.730197270711262, 35.22677397727966]]
MS-SSIM: [[0.6105455855528513, 0.8047609776258469, 0.8287305037180582, 0.8523357113202413, 0.9162356704473495], [0.6661650314927101, 0.8659089903036753, 0.8838321218887965, 0.920901820063591, 0.9688620070616404], [0.694376602768898, 0.8907284761468569, 0.9031753142674764, 0.9405034507314364, 0.9821811690926552]]
Finish Test!

画出来的图像是

4. 与您在论文图9的图像略有差距，在信道比=0.02的时候psnr是十几，和曲线上升不平滑。

请问，是我哪里操作错误了吗？还是选错模型啦

[AbdulrahmanSoliman1]

Same issue

[semcomm]

抱歉，隔了这么久才看到消息，我刚刚按照您的指令测了一下，得出来的结果是这个，我也不知道您哪里出现了问题
SNR: [[0.0, 0.0, 0.0, 0.0, 0.0], [4.0, 4.0, 4.0, 4.0, 4.0], [8.0, 8.0, 8.0, 8.0, 8.0]]
CBR: [[0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125], [0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125], [0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125]]
PSNR: [[25.62018044789632, 27.288144826889038, 28.224758704503376, 28.87304186820984, 29.695185979207356], [27.65361976623535, 29.732901970545452, 31.089426676432293, 31.979695161183674, 33.185670137405396], [28.700441598892212, 31.077546675999958, 32.602484385172524, 33.72682213783264, 35.21999963124593]]

[AbdulrahmanSoliman1]

Sorry, it took so long to see the message, I just tested it according to your instructions, and the result is this, and I don't know where you have the problem SNR: [[0.0, 0.0, 0.0, 0.0, 0.0], [4.0, 4.0, 4.0, 4.0, 4.0], [8.0, 8.0, 8.0, 8.0, 8.0]]. CBR: [[0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125], [0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125], [ 0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125]] PSNR: [[25.62018044789632, 27.288144826889038, 28.224758704503376, 28.87304186820984, 29.695185979207356], [27.65361976623535, 29.732901970545452, 31.089426676432293, 31.979695161183674, 33.185670137405396], [28.700441598892212, 31.077546675999958, 32.602484385172524, 33.72682213783264, 35.21999963124593]]

can you give use the command you used and which model

[homesomele]

你好，请问你加载自己训练出来的模型会不会遇到报错呀，我一直显示模型参数不匹配
Traceback (most recent call last):
File "/home/yiyicai/SenCom/SwinJSCC-main/test.py", line 354, in
load_weights(model_path)
File "/home/yiyicai/SenCom/SwinJSCC-main/test.py", line 155, in load_weights
net.load_state_dict(pretrained, strict=False)
File "/home/yiyicai/miniconda3/envs/mask/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1604, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for SwinJSCC:
size mismatch for encoder.layers.0.blocks.1.attn_mask: copying a param with shape torch.Size([1536, 64, 64]) from checkpoint, the shape in current model is torch.Size([256, 64, 64]).
size mismatch for encoder.layers.1.blocks.1.attn_mask: copying a param with shape torch.Size([384, 64, 64]) from checkpoint, the shape in current model is torch.Size([64, 64, 64]).
size mismatch for encoder.layers.2.blocks.1.attn_mask: copying a param with shape torch.Size([96, 64, 64]) from checkpoint, the shape in current model is torch.Size([16, 64, 64]).
size mismatch for encoder.layers.2.blocks.3.attn_mask: copying a param with shape torch.Size([96, 64, 64]) from checkpoint, the shape in current model is torch.Size([16, 64, 64]).
size mismatch for encoder.layers.2.blocks.5.attn_mask: copying a param with shape torch.Size([96, 64, 64]) from checkpoint, the shape in current model is torch.Size([16, 64, 64]).
size mismatch for encoder.layers.3.blocks.1.attn_mask: copying a param with shape torch.Size([24, 64, 64]) from checkpoint, the shape in current model is torch.Size([4, 64, 64]).
size mismatch for decoder.layers.0.blocks.1.attn_mask: copying a param with shape torch.Size([24, 64, 64]) from checkpoint, the shape in current model is torch.Size([4, 64, 64]).
size mismatch for decoder.layers.1.blocks.1.attn_mask: copying a param with shape torch.Size([96, 64, 64]) from checkpoint, the shape in current model is torch.Size([16, 64, 64]).
size mismatch for decoder.layers.1.blocks.3.attn_mask: copying a param with shape torch.Size([96, 64, 64]) from checkpoint, the shape in current model is torch.Size([16, 64, 64]).
size mismatch for decoder.layers.1.blocks.5.attn_mask: copying a param with shape torch.Size([96, 64, 64]) from checkpoint, the shape in current model is torch.Size([16, 64, 64]).
size mismatch for decoder.layers.2.blocks.1.attn_mask: copying a param with shape torch.Size([384, 64, 64]) from checkpoint, the shape in current model is torch.Size([64, 64, 64]).
size mismatch for decoder.layers.3.blocks.1.attn_mask: copying a param with shape torch.Size([1536, 64, 64]) from checkpoint, the shape in current model is torch.Size([256, 64, 64]).

[Jeffry1997]

Sorry, it took so long to see the message, I just tested it according to your instructions, and the result is this, and I don't know where you have the problem SNR: [[0.0, 0.0, 0.0, 0.0, 0.0], [4.0, 4.0, 4.0, 4.0, 4.0], [8.0, 8.0, 8.0, 8.0, 8.0]]. CBR: [[0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125], [0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125], [ 0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125]] PSNR: [[25.62018044789632, 27.288144826889038, 28.224758704503376, 28.87304186820984, 29.695185979207356], [27.65361976623535, 29.732901970545452, 31.089426676432293, 31.979695161183674, 33.185670137405396], [28.700441598892212, 31.077546675999958, 32.602484385172524, 33.72682213783264, 35.21999963124593]]

can you give use the command you used and which model

非常感谢您的帮忙，已经运行起来啦。但是我又遇见一个小问题 1.加载预训练模型：model_path=("./SwinJSCC/SwinJSCC w- SA&RA/SwinJSCC_w_SAandRA_AWGN_HRimage_cbr_psnr_snr.model")

2.测试命令：--trainset DIV2K --testset kodak --distortion-metric MSE --model SwinJSCC_w/_SAandRA --channel-type awgn --C 32,64,96,128,192 --multiple-snr 0,4,8 --model_size base

3.出现的结果是SNR: [[0.0, 0.0, 0.0, 0.0, 0.0], [4.0, 4.0, 4.0, 4.0, 4.0], [8.0, 8.0, 8.0, 8.0, 8.0]] CBR: [[0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125], [0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125], [0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125]] PSNR: [[20.404406030972797, 25.127057949701946, 25.701947053273518, 26.854474862416584, 29.68847068150838], [20.880515893300373, 26.36423667271932, 26.87330198287964, 28.887059609095257, 33.189711252848305], [21.143070936203003, 26.90866978963216, 27.359466234842937, 29.730197270711262, 35.22677397727966]] MS-SSIM: [[0.6105455855528513, 0.8047609776258469, 0.8287305037180582, 0.8523357113202413, 0.9162356704473495], [0.6661650314927101, 0.8659089903036753, 0.8838321218887965, 0.920901820063591, 0.9688620070616404], [0.694376602768898, 0.8907284761468569, 0.9031753142674764, 0.9405034507314364, 0.9821811690926552]] Finish Test!

画出来的图像是

4. 与您在论文图9的图像略有差距，在信道比=0.02的时候psnr是十几，和曲线上升不平滑。

请问，是我哪里操作错误了吗？还是选错模型啦

您好 解决了吗？我也遇到同样的问题。

[semcomm]

我重新上传了模型SwinJSCC_w_SAandRA_AWGN_HRimage_cbr_psnr_snr.model 希望能给您们提供帮助

[waaanp]

非常感谢您的帮忙，已经运行起来啦。但是我又遇见一个小问题 1.加载预训练模型：model_path=("./SwinJSCC/SwinJSCC w- SA&RA/SwinJSCC_w_SAandRA_AWGN_HRimage_cbr_psnr_snr.model")

2.测试命令：--trainset DIV2K --testset kodak --distortion-metric MSE --model SwinJSCC_w/_SAandRA --channel-type awgn --C 32,64,96,128,192 --multiple-snr 0,4,8 --model_size base

3.出现的结果是SNR: [[0.0, 0.0, 0.0, 0.0, 0.0], [4.0, 4.0, 4.0, 4.0, 4.0], [8.0, 8.0, 8.0, 8.0, 8.0]] CBR: [[0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125], [0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125], [0.020833333333333325, 0.04166666666666665, 0.0625, 0.0833333333333333, 0.125]] PSNR: [[20.404406030972797, 25.127057949701946, 25.701947053273518, 26.854474862416584, 29.68847068150838], [20.880515893300373, 26.36423667271932, 26.87330198287964, 28.887059609095257, 33.189711252848305], [21.143070936203003, 26.90866978963216, 27.359466234842937, 29.730197270711262, 35.22677397727966]] MS-SSIM: [[0.6105455855528513, 0.8047609776258469, 0.8287305037180582, 0.8523357113202413, 0.9162356704473495], [0.6661650314927101, 0.8659089903036753, 0.8838321218887965, 0.920901820063591, 0.9688620070616404], [0.694376602768898, 0.8907284761468569, 0.9031753142674764, 0.9405034507314364, 0.9821811690926552]] Finish Test!

画出来的图像是

4. 与您在论文图9的图像略有差距，在信道比=0.02的时候psnr是十几，和曲线上升不平滑。

请问，是我哪里操作错误了吗？还是选错模型啦

您好 最后代码复现了吗