deepliif 1.2.3__py3-none-any.whl → 1.2.4__py3-none-any.whl

deepliif/util/__init__.py

@@ -377,6 +377,104 @@ def adjust_background_tile(img):
     return image
 
 
+def infer_background_colors(dir_data, sample_size=5, input_no=1, modalities_no=4,
+                                    seg_no=1, tile_size=32, return_list=False):
+    fns = [x for x in os.listdir(dir_data) if x.endswith('.png')]
+    sample_size = min(sample_size, len(fns))
+    w, h, num_img = None, None, None
+
+    background_colors = {}
+
+    count = 0
+    while count < sample_size and len(fns) > 0: 
+        fn = fns.pop(0)
+        img = Image.open(f"{dir_data}/{fn}")
+        
+        if w is None:
+            num_img = img.size[0] / img.size[1]
+            num_img = int(num_img)
+            w, h = img.size
+        
+        background_colors_img = infer_background_colors_for_img(img, input_no=input_no, modalities_no=modalities_no, seg_no=seg_no, tile_size=tile_size, w=w, h=h, num_img=num_img)
+        
+        if background_colors_img is not None:
+            count += 1
+            for mod_id, rgb_avg in background_colors_img.items():
+                try:
+                    background_colors[mod_id].append(rgb_avg)
+                except:
+                    background_colors[mod_id] = [rgb_avg]
+    
+    if count > 0:
+        print(f'Calculating average color for empty tiles from {count} images..')
+        background_colors = {k:np.mean(v,axis=0).astype(np.uint8) for k,v in background_colors.items()}
+        
+        if return_list:
+            return [tuple(e) for e in background_colors.values()]
+        else:
+            return background_colors
+    else:
+        print('None of the images have empty tiles for estimating averge background color. Try with a proper tile size.')
+        return None
+
+
+def infer_background_colors_for_img(img, input_no=1, modalities_no=4, seg_no=1, tile_size=32,
+                                    w=None, h=None, num_img=None):
+    """
+    Estimate background colors for a given RGB image.
+    The empty tiles are determined by applying is_empty() function on segmentation modalities.
+    If multiple segmentation modalities present, only common empty tiles are used for background
+    color calculation.
+    """
+    from ..models import is_empty
+    
+    if w is None:
+        num_img = img.size[0] / img.size[1]
+        num_img = int(num_img)
+        w, h = img.size
+            
+    empty_tiles = {}
+    l_box = []
+    background_colors = {}
+    
+    for i in range(num_img-seg_no, num_img):
+        img_mod = img.crop((h*i,0,h*(i+1),h))
+        l_box_mod = []
+        for x in range(0, h, tile_size):
+            for y in range(0, h, tile_size):
+                box = (x, y, x+tile_size, y+tile_size)
+                tile = img_mod.crop(box)
+                if is_empty(tile):
+                    l_box_mod.append(box)
+        l_box.append(l_box_mod)
+
+    l_box_final = set()
+    if len(l_box) > 1:
+        # only keep overlapped boxes
+        for l in l_box:
+            l_box_final = l_box_final & set(l)
+        l_box_final = list(l_box_final)
+    else:
+        l_box_final = l_box[0]
+    #print(f'{len(l_box_final)} tiles are considered empty using segmentation modalities')
+    
+    if len(l_box_final) == 0: # this can happen for images with dense cells
+        return None
+
+    for i in range(input_no, modalities_no+input_no):
+        empty_tiles[i] = []
+        img_mod = img.crop((h*i,0,h*(i+1),h))
+        for box in l_box_final:
+            tile = img_mod.crop(box)
+            empty_tiles[i].append(tile)
+
+        img_avg = np.mean(np.stack(empty_tiles[i], axis=0), axis=0) # take an average across all empty images
+        rgb_avg = np.mean(img_avg,axis=(0,1)).astype(np.uint8)
+        background_colors[i] = rgb_avg
+
+    return background_colors
+
+
 def image_variance_gray(img):
     px = np.asarray(img) if img.mode == 'L' else np.asarray(img.convert('L'))
     idx = np.logical_and(px != 255, px != 0)
@@ -531,7 +629,6 @@ class WSIReader:
 
 
 
-
 def write_results_to_pickle_file(output_addr, results):
     """
     This function writes data into the pickle file.

deepliif/util/util.py

@@ -205,3 +205,88 @@ class HardwareStatus():
         self.timer.cancel()
 
 
+def get_mod_id_seg(dir_model):
+    # assume we already know there are seg models - this check is intended to be done prior to calling this function
+    fns = [fn for fn in os.listdir(dir_model) if fn.endswith('.pth') and 'net_G' in fn]
+    
+    if len(fns) == 0: # typically this means the directory only contains serialized models
+        fns = [fn for fn in os.listdir(dir_model) if fn.endswith('.pt') and fn.startswith('G')]
+        model_names = [fn[1:-3] for fn in fns] # 1[:-3] drop ".pt" and the starting G
+    else:
+        model_names = [fn[:-4].split('_')[2][1:] for fn in fns] # [1:] drop "G"
+    
+    if len(fns) == 0:
+        raise Exception('Cannot find any model file ending with .pt or .pth in directory',dir_model)
+    
+    model_name_seg = max(model_names, key=len)
+    return model_name_seg[0]
+
+def get_input_id(dir_model):
+    # assume we already know there are seg models - this check is intended to be done prior to calling this function
+    fns = [fn for fn in os.listdir(dir_model) if fn.endswith('.pth') and 'net_G' in fn]
+    
+    if len(fns) == 0: # typically this means the directory only contains serialized models
+        fns = [fn for fn in os.listdir(dir_model) if fn.endswith('.pt') and fn.startswith('G')]
+        model_names_seg = [fn[2:-3] for fn in fns] # [2:] drop "GS"/"G5"
+    else:
+        model_names_seg = [fn[:-4].split('_')[2][2:] for fn in fns] # [2:] drop "GS"/"G5"
+    
+    if len(fns) == 0:
+        raise Exception('Cannot find any model file ending with .pt or .pth in directory',dir_model)
+      
+    if '0' in model_names_seg:
+        return '0'
+    else:
+        return '1'
+    
+def init_input_and_mod_id(opt, dir_model=None):
+    """
+    Used by model classes to initialize input id and mod id under different situations.
+    """
+    mod_id_seg = None
+    input_id = None
+    
+    if not opt.continue_train and opt.is_train:
+        if hasattr(opt,'mod_id_seg'): # use mod id seg from train opt file if available
+            mod_id_seg = opt.mod_id_seg
+        elif not hasattr(opt,'modalities_names'): # backward compatible with models trained before this param was introduced
+            mod_id_seg = opt.modalities_no + 1 # for original DeepLIIF, modalities_no is 4 and the seg mod id is 5
+        else:
+            mod_id_seg = 'S'
+        
+        if opt.model in ['DeepLIIF','DeepLIIFKD']:
+            input_id = '0'
+    else: # for both continue train and test, we load existing models, so need to obtain seg mod id and input id from filenames
+        if hasattr(opt, 'mod_id_seg'):
+            mod_id_seg = opt.mod_id_seg
+        else:
+            # for contiue-training, extract mod id seg from existing files if not available
+            mod_id_seg = get_mod_id_seg(dir_model if dir_model is not None else os.path.join(opt.checkpoints_dir, opt.name))
+        
+        if opt.model in ['DeepLIIF','DeepLIIFKD']:
+            input_id = get_input_id(dir_model if dir_model is not None else os.path.join(opt.checkpoints_dir, opt.name))
+        
+    return mod_id_seg, input_id
+
+
+import copy
+def map_model_names(model_names,mod_id_seg_source,input_id_source,mod_id_seg_target,input_id_target):
+    """
+    Used for DeepLIIFKD to map model/image names from teacher model to student model,
+    when mod_id_seg and/or input_id is different.
+    """
+    d_res = {}
+    for model_name in model_names:
+        model_name_new = copy.deepcopy(model_name)
+        if len(model_name) > 2 and model_name[1] == mod_id_seg_source:
+            model_name_new = model_name[0]+mod_id_seg_target+model_name[2:] # replace mod id seg
+            if input_id_source != input_id_target: # currently only support 0 or 1 as input_id
+                if int(input_id_target) == 0:
+                    model_name_new = model_name_new[:2] + str(int(model_name_new[2:])-1)
+                else:
+                    model_name_new = model_name_new[:2] + str(int(model_name_new[2:])+1)
+        d_res[model_name] = model_name_new
+    # this is not a model name but the final aggregated segmentation image name
+    # so it cannot be obtained from model names
+    d_res['G'+mod_id_seg_source] = 'G'+mod_id_seg_target
+    return d_res

deepliif/util/visualizer.py

@@ -138,7 +138,7 @@ class Visualizer():
         print('Command: %s' % cmd)
         Popen(cmd, shell=True, stdout=PIPE, stderr=PIPE)
 
-    def display_current_results(self, visuals, epoch, save_result, **kwargs):
+    def display_current_results(self, visuals, epoch, save_result, filename=None, **kwargs):
         """Display current results on visdom; save current results to an HTML file.
 
         Parameters:
@@ -171,7 +171,7 @@ class Visualizer():
                             table td {width: % dpx; height: % dpx; padding: 4px; outline: 4px solid black}
                             </style>""" % (w, h)  # create a table css
                     # create a table of images.
-                    title = self.name
+                    title = self.name if filename is None else filename
                     label_html = ''
                     label_html_row = ''
                     images = []

{deepliif-1.2.3.dist-info → deepliif-1.2.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: deepliif
-Version: 1.2.3
+Version: 1.2.4
 Summary: DeepLIIF: Deep-Learning Inferred Multiplex Immunofluorescence for Immunohistochemical Image Quantification
 Home-page: https://github.com/nadeemlab/DeepLIIF
 Author: Parmida93
@@ -66,7 +66,7 @@ segmentation.*
 
 © This code is made available for non-commercial academic purposes.
 
-![Version](https://img.shields.io/static/v1?label=latest&message=v1.2.3&color=darkgreen)
+![Version](https://img.shields.io/static/v1?label=latest&message=v1.2.4&color=darkgreen)
 [![Total Downloads](https://static.pepy.tech/personalized-badge/deepliif?period=total&units=international_system&left_color=grey&right_color=blue&left_text=total%20downloads)](https://pepy.tech/project/deepliif?&left_text=totalusers)
 
 ![overview_image](./images/overview.png)*Overview of DeepLIIF pipeline and sample input IHCs (different 

{deepliif-1.2.3.dist-info → deepliif-1.2.4.dist-info}/RECORD

@@ -1,9 +1,9 @@
-cli.py,sha256=luwzcSrRZMePfJ-h4Hgz2X7lYC70TUVxGfEo3LyPJgQ,58938
+cli.py,sha256=hu-_-yO-ytr5gy3_Xd-kdZ7GdSUR_mISci7IE7nDKg0,63092
 deepliif/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-deepliif/postprocessing.py,sha256=Uu4axoOLgPgBgXxSVAgviuLi9cr4U_eE4KgFLbleMro,43512
+deepliif/postprocessing.py,sha256=zMYlUeoz1tU7Cyd9VIxFez126CxBEcc1eBf6nFx-4Rg,43512
 deepliif/data/__init__.py,sha256=IfqVFnFSPQJZnORdRq4sNkJiylr1TaKNmhvWP_aLHdg,5492
 deepliif/data/aligned_dataset.py,sha256=CkKXj94ANSi8RdhpRQjVETFYlRMER2XErIf-87BStTE,5175
-deepliif/data/base_dataset.py,sha256=bQlxfY7bGSE9WPj31ZHkCxv5CAEJovjakGDCcK-aYdc,5564
+deepliif/data/base_dataset.py,sha256=6ZEjIHDJ7f2JdjxZjbxu9QWnQDoIDmUQw4yPvLRmB2Y,5661
 deepliif/data/colorization_dataset.py,sha256=uDYWciSxwqZkStQ_Vte27D9x5FNhv6eR9wSPn39K3RY,2808
 deepliif/data/image_folder.py,sha256=eesP9vn__YQ-dw1KJG9J-yVUHMmJjLcIEQI552Iv2vE,2006
 deepliif/data/single_dataset.py,sha256=hWjqTkRESEMppZj_r8bi3G0hAZ5EfvXYgE_qRbpiEz4,1553
@@ -11,28 +11,28 @@ deepliif/data/template_dataset.py,sha256=PCDBnFRzRKReaeWgKUZmW0LrzRByI9adrKDJ6SN
 deepliif/data/unaligned_dataset.py,sha256=D69SxV087jKTd990UQIR4F3TahJHiSiw8i9Uz_xybt0,4697
 deepliif/models/CycleGAN_model.py,sha256=WDEa-Zgz57mVc9HbcVDXL5vfHvUDWdWXNLyz8ReH3rg,15196
 deepliif/models/DeepLIIFExt_model.py,sha256=HZaX9Z2ue0HQCFFN3guLkBcByCP70i8JvmPY02oOMyU,15022
-deepliif/models/DeepLIIFKD_model.py,sha256=edq9fxrDspGivuFlAYZp9B0Opp3BRIosA9e1TI_gxpc,27152
-deepliif/models/DeepLIIF_model.py,sha256=6vmsXcBcoALrhJLa7XGhDmLamO_WCzTDYEyVUBE482o,23857
+deepliif/models/DeepLIIFKD_model.py,sha256=YMJrun6cGLyukRj4u3yvgm7Z1IOG1juxT11eyX25BRM,25985
+deepliif/models/DeepLIIF_model.py,sha256=hMXE_6NuZvE8UxMrDSNGMFi_zBlW5VqBGj05qzSc5ec,30296
 deepliif/models/SDG_model.py,sha256=3opz7uEyhvVJ8fF4_Jw4ho1MBcc9OVye-ByZD_KF2j0,10142
-deepliif/models/__init__.py,sha256=FLMYnTyF4XvsSqN11df4a7dqFuB8-Nq4pQnTskL2DUE,34398
+deepliif/models/__init__.py,sha256=l6z3aZneEA6ZwrmwDb4sS-FtKSntYNQJMVX0UNF0qTE,40409
 deepliif/models/att_unet.py,sha256=tqaFMNbGQUjXObOG309P76c7sIPxEvFR38EyuyHY40o,7116
-deepliif/models/base_model.py,sha256=ZQBI-wVfDdu296HzB_YzQraE9oUwfyRlAolNlrMi-4g,16858
+deepliif/models/base_model.py,sha256=PNv_JwFQ1_u4MUfqTWd8IXr1JHf5_vLkCnIU3ePJftg,17116
 deepliif/models/networks.py,sha256=SclYeZ78U9ZSOhNrky764ZXB6muIqnHa6K0h_LDSCi0,36690
-deepliif/options/__init__.py,sha256=p2IWckf3-K-wclDWfSq5ZmynKk2lNov2Tn7WPYIO11A,8329
+deepliif/options/__init__.py,sha256=TIPw7LuwvZ8yUTMBZNf3aLKM_jOiD9_REUx102oWjiQ,10336
 deepliif/options/base_options.py,sha256=m5UXY8MvjNcDisUWuiP228yoT27SsCh1bXS_Td6SwTc,9852
 deepliif/options/processing_options.py,sha256=OnNT-ytoTQzetFiMEKrWvrsrhZlupRK4smcnIk0MbqY,2947
 deepliif/options/test_options.py,sha256=4ZbQC5U-nTbUz8jvdDIbse5TK_mjw4D5yNjpVevWD5M,1114
 deepliif/options/train_options.py,sha256=5eA_oxpRj2-HiuMMvC5-HLapxNFG_JXOQ3K132JjpR8,3580
-deepliif/util/__init__.py,sha256=Vl_mUSCFLWZiX-eN9I-DBrbJlW29G8nPo8b5Je5pHnY,31582
+deepliif/util/__init__.py,sha256=iqC3FvU7QDHdChZtf-nAaUzO61y02a7BiDmxpL679tc,35302
 deepliif/util/checks.py,sha256=xQirKbZxZErsAXc27M5miQUUyxptoIEJSDaUKv1nY7c,1371
 deepliif/util/get_data.py,sha256=HaRoQYb2u0LUgLT7ES-w35AmJ4BrlBEJWU4Cok29pxI,3749
 deepliif/util/html.py,sha256=RNAONZ4opP-bViahgmpSbHwOc6jXKQRnWRAVIaeIvac,3309
 deepliif/util/image_pool.py,sha256=M89Hc7DblRWroNP71S9mAdRn7h3DrhPFPjqFxxZYSgw,2280
-deepliif/util/util.py,sha256=WV2a1Rt-McmZm3BKW7TqC2oAKkvxWhZvjofGSfY6y7s,6810
-deepliif/util/visualizer.py,sha256=6E1sPbXdgLFB9mnPwtfEjm9O40viG4dfv5MyTpOQQpo,20210
-deepliif-1.2.3.dist-info/LICENSE.md,sha256=HlZw_UPS6EtJimJ_Ci7xKh-S5Iubs0Z8y8E6EZ3ZNyE,956
-deepliif-1.2.3.dist-info/METADATA,sha256=oQZ1gLKnysSKLZB98HB71yn-wAY8KWnpHb5uyOP2kbk,35266
-deepliif-1.2.3.dist-info/WHEEL,sha256=pkctZYzUS4AYVn6dJ-7367OJZivF2e8RA9b_ZBjif18,92
-deepliif-1.2.3.dist-info/entry_points.txt,sha256=f70-10j2q68o_rDlsE3hspnv4ejlDnXwwGZ9JJ-3yF4,37
-deepliif-1.2.3.dist-info/top_level.txt,sha256=vLDK5YKmDz08E7PywuvEjAo7dM5rnIpsjR4c0ubQCnc,13
-deepliif-1.2.3.dist-info/RECORD,,
+deepliif/util/util.py,sha256=l7QsUL8gSMJSGH5-Nin6IsjPl9HTyaKk8Z5cd-yuPjs,11055
+deepliif/util/visualizer.py,sha256=BeRIEzbpx0YCn1_zFCtV1JHLMzyWed_nw65wiyKRlcc,20259
+deepliif-1.2.4.dist-info/LICENSE.md,sha256=HlZw_UPS6EtJimJ_Ci7xKh-S5Iubs0Z8y8E6EZ3ZNyE,956
+deepliif-1.2.4.dist-info/METADATA,sha256=dscN8jbrXPjEWzgs3q27L24ft2vqiDXiCcDoIKoN1Ac,35266
+deepliif-1.2.4.dist-info/WHEEL,sha256=pkctZYzUS4AYVn6dJ-7367OJZivF2e8RA9b_ZBjif18,92
+deepliif-1.2.4.dist-info/entry_points.txt,sha256=f70-10j2q68o_rDlsE3hspnv4ejlDnXwwGZ9JJ-3yF4,37
+deepliif-1.2.4.dist-info/top_level.txt,sha256=vLDK5YKmDz08E7PywuvEjAo7dM5rnIpsjR4c0ubQCnc,13
+deepliif-1.2.4.dist-info/RECORD,,