add gap tv+ffdnet and gaptv+fastdvdnet

Author: dawnlh

Diff for: PnP_SCI/matlab/results/savedmat/combine_binary_mask_256_10f/test_performance.txt

@@ -13,5 +13,5 @@ ssim	0.6603	0.6930	0.8944	0.6190	0.7439	0.4921	0.6838
 
 * GAP-TV+FFDNET
 		aerial	crash	drop	kobe	runner	traffic	[average]
-psnr	22.34	22.65	31.40	24.41	26.65	19.11	24.4267
+psnr	22.34	22.65	31.40	24.88	26.65	19.11	24.4267
 ssim									0.8730	0.6047

Diff for: PnP_SCI/matlab/test_pnpsci_meas_test.m

@@ -35,7 +35,7 @@
 % datasetdir = './dataset/simdata/cacti/cacti_256_10f_1';
 resultdir  = './results';                   % results
 
-test_algo_flag = [1, 4];		% choose algorithms: 0-all, 1-gaptv, 2-gap-ffdnet, 3-ista-tv, 4-gap-tv+ffdnet
+test_algo_flag = [1,4];		% choose algorithms: 0-all, 1-gaptv, 2-gap-ffdnet, 3-ista-tv, 4-gap-tv+ffdnet
 saving_data_flag = 0;	% save the recon result
 tv_init_flag = 0;		% use gap-tv recon as initial image for gap-ffdnet
 show_res_flag = 0;
@@ -96,12 +96,12 @@
 if ismember(0,test_algo_flag) || ismember(1,test_algo_flag)
 	para.denoiser = 'tv'; % TV denoising
 % 	para.tvm = 'ITV3D_FGP';  % tv denoiser
-	para.tvm = 'original';  % tv denoiser
+	para.tvm = 'ATV_ClipA';  % tv denoiser
 	para.maxiter  = 100; % maximum iteration
 	% para.tvweight = 0.07*255/MAXB; % weight for TV denoising, original
 	% para.tviter   = 5; % number of iteration for TV denoising, original
 
-	para.tvweight = 0.07*255/MAXB; % weight for TV denoising, test
+	para.tvweight = 0.15*255/MAXB; % weight for TV denoising, test
 	para.tviter   = 5; % number of iteration for TV denoising, test
 
 	[vgaptv,psnr_gaptv,ssim_gaptv,tgaptv,psnrall_gaptv] = ...

Diff for: PnP_SCI/python/dvp_linear_inv.py

@@ -441,6 +441,7 @@ def admm_denoise_bayer(y_bayer, Phi_bayer, _lambda=1, gamma=0.01,
     return x_bayer, psnr_all
 
 
+# admm(gap) denosie cacti (including gap)
 def admmdenoise_cacti(meas, mask, A, At, projmeth='admm', v0=None, orig=None, 
                       iframe=0, nframe=1, MAXB=1., maskdirection='plain',
                       **args):
@@ -457,7 +458,7 @@ def admmdenoise_cacti(meas, mask, A, At, projmeth='admm', v0=None, orig=None,
     begin_time = time.time()
     # loop over all the coded frames [nframe]
     for kf in range(nframe):
-        print('%s-%s Reconstruction coded frame block %2d of %2d ...'
+        print('\n=== %s-%s Reconstruction coded frame block %2d of %2d ==='
               %(projmeth.upper(), args['denoiser'].upper(), kf+1, nframe))
         if orig is not None:
             orig_k = orig[:,:,(kf+iframe)*nmask:(kf+iframe+1)*nmask]/MAXB

Diff for: PnP_SCI/python/info_log.txt

@@ -0,0 +1,3 @@
+info_log:
+conda environment:
+    windows: dvp

Diff for: PnP_SCI/python/joint_dvp_linear_inv.py

@@ -0,0 +1,55 @@
+# combine_binary_mask
+* use optimal parametes for kobe
+
+bm_256_10f:
+---------
+GAP-TV
+		aerial	crash	drop	kobe	runner	traffic	[average]
+psnr	22.33	21.88	24.82	24.09	25.71	18.61	22.9067
+ssim	0.7504	0.7609	0.9046	0.7655	0.8556	0.5896	0.7711
+
+
+
+---------
+GAP-TV+FFDNET
+		aerial	crash	drop	kobe	runner	traffic	[average]
+psnr	22.87	23.14	28.96	25.24	26.97	19.28	24.41
+ssim	0.7325	0.7144	0.9233	0.7528	0.8036	0.5858	0.7521
+
+
+
+
+---------
+GAP-TV+FASTDVDNET
+		aerial	crash	drop	kobe	runner	traffic	[average]
+psnr	25.55	25.83	36.03	27.32	32.10	21.77	28.10
+ssim	0.8258	0.8607	0.9547	0.8583	0.9263	0.7492	0.8625
+
+
+
+
+
+param log：
+GAP-TV：
+    projmeth = 'gap' # projection method
+    _lambda = 1 # regularization factor, [original set]
+    accelerate = True # enable accelerated version of GAP
+    denoiser = 'tv' # total variation (TV)
+    iter_max = 100 # maximum number of iterations
+    tv_weight = 0.25 # TV denoising weight (larger for smoother but slower) [kobe:0.25; ]
+    tv_iter_max = 5 # TV denoising maximum number of iterations each
+
+GAP-TV+FASTDVDNET
+    projmeth = 'gap' # projection method
+    _lambda = 1 # regularization factor, [original set]
+    accelerate = True # enable accelerated version of GAP
+    denoiser = 'tv+fastdvdnet' # video non-local network 
+    noise_estimate = False # disable noise estimation for GAP
+    sigma1    = None # pre-set noise standard deviation for 1st period denoise 
+    iter_max1 = 100 # maximum number of iterations for 1st period denoise   
+    sigma2    = [100/255, 50/255, 25/255] # pre-set noise standard deviation for 2nd period denoise 
+    iter_max2 = [80, 100, 150] # maximum number of iterations for 2nd period denoise    
+    # sigma2    = [50/255, 25/255] # pre-set noise standard deviation for 2nd period denoise 
+    # iter_max2 = [20, 20] # maximum number of iterations for 2nd period denoise   
+    tv_iter_max = 5 # TV denoising maximum number of iterations each
+    tv_weight = 0.25 # TV denoising weight (larger for smoother but slower) [kobe:0.25]

Diff for: PnP_SCI/python/pnp_sci_video_meas_test.py

@@ -0,0 +1,53 @@
+# combine_binary_mask_256_10f_2_uniform
+* use optimal parametes for each data
+
+bm_256_10f:
+---------
+GAP-TV
+		aerial	crash	drop	kobe	runner	traffic	[average]
+psnr	24.05	24.14	31.85	25.35	28.27	19.88	25.59
+ssim	0.7857	0.7953	0.9465	0.8016	0.8818	0.6438	0.8091
+
+
+
+---------
+GAP-TV+FFDNET
+		aerial	crash	drop	kobe	runner	traffic	[average]
+psnr	
+ssim	
+
+
+
+
+---------
+GAP-TV+FASTDVDNET
+		aerial	crash	drop	kobe	runner	traffic	[average]
+psnr	26.33	26.79	38.39	28.13	32.58	22.73	29.1583
+ssim	0.8524	0.8889	0.9821	0.8866	0.9402	0.778	0.8880
+
+
+
+param log：
+GAP-TV：
+    projmeth = 'gap' # projection method
+    _lambda = 1 # regularization factor, [original set]
+    accelerate = True # enable accelerated version of GAP
+    denoiser = 'tv' # total variation (TV)
+    iter_max = 100 # maximum number of iterations
+    tv_weight = 0.25 # TV denoising weight (larger for smoother but slower) [kobe:0.25; ]
+    tv_iter_max = 5 # TV denoising maximum number of iterations each
+
+GAP-TV+FASTDVDNET
+    projmeth = 'gap' # projection method
+    _lambda = 1 # regularization factor, [original set]
+    accelerate = True # enable accelerated version of GAP
+    denoiser = 'tv+fastdvdnet' # video non-local network 
+    noise_estimate = False # disable noise estimation for GAP
+    sigma1    = None # pre-set noise standard deviation for 1st period denoise 
+    iter_max1 = 100 # maximum number of iterations for 1st period denoise   
+    sigma2    = [100/255, 50/255, 25/255] # pre-set noise standard deviation for 2nd period denoise 
+    iter_max2 = [60, 100, 150] # maximum number of iterations for 2nd period denoise     
+    # sigma2    = [50/255, 25/255] # pre-set noise standard deviation for 2nd period denoise 
+    # iter_max2 = [20, 20] # maximum number of iterations for 2nd period denoise   
+    tv_iter_max = 5 # TV denoising maximum number of iterations each
+    tv_weight = 0.25 # TV denoising weight (larger for smoother but slower) [kobe:0.25]

Diff for: PnP_SCI/python/pnp_sci_video_orig_test.py

@@ -0,0 +1,7 @@
+## tv denoisers
+
+def denoise_tv_cham_ITV2D():
+    pass
+
+def denoise_tv_FGP_ITV3D():
+    pass

Diff for: PnP_SCI/python/results/savedmat/cacti/combine_binary_mask_256_10f/bm_256_10f/admmfastdvdnet_runner_Cr10_orig.png

@@ -1,6 +1,11 @@
 ''' Utilities '''
 import math
 import numpy as np
+import matplotlib.pyplot as plt
+from statistics import mean
+import scipy.io as sio
+import os
+
 
 def A_(x, Phi):
     '''
@@ -29,3 +34,64 @@ def psnr(ref, img):
         return 100
     PIXEL_MAX = 1.
     return 20 * math.log10(PIXEL_MAX / math.sqrt(mse))
+
+
+
+## zzh
+def show_n_save_res(vdenoise,tdenoise,psnr_denoise,ssim_denoise,psnrall_denoise, orig, Cr, resultsdir, 
+                    save_name, MAXB=255, show_res_flag=1, save_res_flag=1):
+    # show res
+    if show_res_flag:
+        plt.ion()
+        fig = plt.figure(figsize=(12, 6.5))
+        row_num = 5
+        savedfigdir = resultsdir + '/savedfig/cacti/'
+        
+        for nt in range(Cr):
+            plt.subplot(Cr//row_num, row_num, nt+1)
+            plt.imshow(orig[:,:,nt]/MAXB, cmap=plt.cm.gray, vmin=0, vmax=1)
+            plt.axis('off')
+            plt.title('Ground truth: Frame #{0:d}'.format(nt+1), fontsize=12)
+        plt.subplots_adjust(wspace=0.02, hspace=0.02, bottom=0, top=1, left=0, right=1)
+        plt.savefig('{}{}_Cr{:d}_orig.png'.format(savedfigdir,save_name,Cr)) 
+
+        fig = plt.figure(figsize=(12, 6.5))
+        PSNR_rec = np.zeros(Cr)
+        for nt in range(Cr):
+            plt.subplot(Cr//row_num, row_num, nt+1)
+            plt.imshow(vdenoise[:,:,nt], cmap=plt.cm.gray, vmin=0, vmax=1)
+            plt.axis('off')
+            plt.title('Frame #{0:d} ({1:2.2f} dB)'.format(nt+1,psnr_denoise[nt]), fontsize=12)
+            
+        # print('Mean PSNR {:2.2f} dB.'.format(mean(psnr_denoise)))
+        # plt.title('-{} mean PSNR {:2.2f} dB'.format(denoiser.upper(),np.mean(PSNR_rec)))
+        plt.subplots_adjust(wspace=0.02, hspace=0.02, bottom=0, top=1, left=0, right=1)
+        plt.savefig('{}{}_Cr{:d}_vdenoise.png'.format(savedfigdir,save_name,Cr)) 
+
+
+        plt.figure()
+        # plt.rcParams["font.family"] = 'monospace'
+        # plt.rcParams["font.size"] = "20"
+        plt.plot(psnr_denoise)
+        # plt.plot(psnr_denoise,color='black')
+        plt.savefig('{}{}_Cr{:d}_psnr_framewise.png'.format(savedfigdir,save_name,Cr)) 
+
+
+        plt.figure()
+        plt.plot(*psnrall_denoise, 'r')
+        plt.savefig('{}{}_Cr{:d}_psnr_all.png'.format(savedfigdir,save_name,Cr)) 
+
+        plt.ioff()
+        
+    # save res
+    if save_res_flag:
+        savedmatdir = resultsdir + '/savedmat/cacti/'
+        if not os.path.exists(savedmatdir):
+            os.makedirs(savedmatdir)
+        sio.savemat('{}{}_Cr{:d}.mat'.format(savedmatdir,save_name,Cr),
+                    {'vdenoise':vdenoise, 
+                    'psnr_denoise':psnr_denoise,
+                    'ssim_denoise':ssim_denoise,
+                    'psnrall_denoise':psnrall_denoise,
+                    'tdenoise':tdenoise
+                    })