Questions about data_generation.py

[lll143653]

1. Is there a busy wait problem? I tested on 12600k-64G and 13900k-64G computers (100 128*128 images), and set the number of threads to 12, but I found that the speed did not seem to be much improved compared to single thread (it took an hour). I tried to replace the asynchronous operation with the thread pool, and the speed was significantly improved (12 threads increased the speed tenfold)
2. Is there a problem with shuffle order? I tried to use the realsr-benchmark dataset for conversion, but when the degenerate type was shuffleorder (eg, line 2), I got the following error.

{
	"name": "ValueError",
	"message": "invalid literal for int() with base 10: 'rnum1'",
	"stack": "---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[3], line 46
     43     # opt['deg_para'] = \"3 720 720/FisrtOrder_Blur_1_1_15_1_2.89_2.09_2.8_1.99_-0.93_2.71_Resize_1_0.49_bicubic_Noise_1_0_0_19.38_0.96_JPEG_1_83.05_SecondOrder_Blur_1_1_15_2_2.18_1.93_2.97_1.77_-1.48_1.7_Resize_1_0.75_area_Noise_1_1_1_22.42_3.06_FinalOrder_Final_1_0_0_0_Resize_area_JPEG_55.38_.png\"
     44     # Process the single image
     45     add_degradation_single(opt)
---> 46 main()

Cell In[3], line 45, in main()
     42 opt['deg_para'] = \"2 6342 6342/shuffleorder_[1, 0, 2, 4, 3, 6, 5]|dppre_rnum1_0.1950097962261581_rnum2_0.5618335821941504_interptype_2|dp0_rnum_0.8760505434779766_ksize_7_theta_2.10432531732895_l1_0_l2_0_l3_1.0345662105098155|dp1_rnum_0.679603296866133_ksize_21_theta_2.111806099334665_l1_0_l2_0_l3_1.8614006353945842|dp2_rnum_0.81675425545463_sf1_0_interptype_2_blursigma_0.8055972814165084        |dp3_interptype_1|dp4_rnum_0.7451448100419809_noiselevel_5_D_0.9576473911564439_0.0_0.0_0.0_0.29499293016618766_0.0_0.0_0.0_0.4873877719609898_U_0.31634729333799017_0.5399692609520926_0.534214943802118_0.6313312564374616_0.17479465123022708_0.4661335686008219_0.7641098206926502_0.7263628347912018_0.33460950990333405|dp5_rnum_0.4343539487950592_jpegfactor_50|dp7_jpegfactor_91\"
     43 # opt['deg_para'] = \"3 720 720/FisrtOrder_Blur_1_1_15_1_2.89_2.09_2.8_1.99_-0.93_2.71_Resize_1_0.49_bicubic_Noise_1_0_0_19.38_0.96_JPEG_1_83.05_SecondOrder_Blur_1_1_15_2_2.18_1.93_2.97_1.77_-1.48_1.7_Resize_1_0.75_area_Noise_1_1_1_22.42_3.06_FinalOrder_Final_1_0_0_0_Resize_area_JPEG_55.38_.png\"
     44 # Process the single image
---> 45 add_degradation_single(opt)

Cell In[3], line 29, in add_degradation_single(opt)
     27     out = add_bsrgan(img, deg_para)
     28 else:
---> 29     out = add_realesrgan(input_image, deg_para)
     31 # Save the degraded image
     32 cv2.imwrite(output_path, out)

File scripts/data_generation/realesrgan.py:21, in add_realesrgan(path, degradation_para)
     17 img_HR = img2tensor(img_HR)
     18 img_HR = torch.unsqueeze(img_HR, 0)
---> 21 img_lq = degradation_realesrgan(img_HR, degradation_para)
     23 image_LR_blur = tensor2img(img_lq[0])
     25 return image_LR_blur

File scripts/data_generation/realesrgan.py:39, in degradation_realesrgan(image, degradation_para)
     36 deg_para = degradation_para.split('_')
     38 #---------------------------- First Order ----------------------------------
---> 39 out = one_order(image, deg_para[0:25], device)
     41 #---------------------------- 4 jpeg parameters ----------------------------
     42 First_jpeg_flag = int(deg_para[23])

File scripts/data_generation/realesrgan.py:145, in one_order(img, Degradation_para, device)
    143 def one_order(img, Degradation_para, device):
--> 145     First_blur_flag = int(Degradation_para[2])
    146     First_blur_para = Degradation_para[3:12]
    148     First_resize_flag = int(Degradation_para[13])

ValueError: invalid literal for int() with base 10: 'rnum1'"
}

and

Error processingdatasets/benchmark_drealsr/test_HR/sony_189_x1.png: Module 'scipy' has no attribute 'finfo'

Is there any potential problem?