Add image crop transform

match_two.py

@@ -95,7 +95,10 @@ def match_two(model, device, config, im_one, im_two, plot_save_path):
 
     model.eval()
 
-    it = input_transform((int(config['feature_extract']['imageresizeH']), int(config['feature_extract']['imageresizeW'])))
+    crop_roi = None
+    if 'imagecrop' in config['feature_extract']:
+        crop_roi = tuple([int(x) for x in config['feature_extract']['imagecrop'].split(",")])
+    it = input_transform((int(config['feature_extract']['imageresizeH']), int(config['feature_extract']['imageresizeW'])), crop_roi)
 
     im_one_pil = Image.fromarray(cv2.cvtColor(im_one, cv2.COLOR_BGR2RGB))
     im_two_pil = Image.fromarray(cv2.cvtColor(im_two, cv2.COLOR_BGR2RGB))
@@ -148,6 +151,10 @@ def match_two(model, device, config, im_one, im_two, plot_save_path):
             tqdm.write('====> Plotting Local Features and save them to ' + str(join(plot_save_path, 'patchMatchings.png')))
 
         # using cv2 for their in-built keypoint correspondence plotting tools
+        if crop_roi is not None:
+            top, left, bottom, right = crop_roi
+            im_one = im_one[top:bottom,left:right]
+            im_two = im_two[top:bottom,left:right]
         cv_im_one = cv2.resize(im_one, (int(config['feature_extract']['imageresizeW']), int(config['feature_extract']['imageresizeH'])))
         cv_im_two = cv2.resize(im_two, (int(config['feature_extract']['imageresizeW']), int(config['feature_extract']['imageresizeH'])))
         # cv2 resize slightly different from torch, but for visualisation only not a big problem

patchnetvlad/tools/datasets.py

@@ -27,6 +27,7 @@
 import os
 
 import torchvision.transforms as transforms
+import torchvision.transforms.functional as TF
 import torch.utils.data as data
 
 import numpy as np
@@ -36,20 +37,19 @@
 from patchnetvlad.tools import PATCHNETVLAD_ROOT_DIR
 
 
-def input_transform(resize=(480, 640)):
+def input_transform(resize=(480, 640), crop_roi=None):
+    trans = []
+    if crop_roi is not None and len(crop_roi) == 4 and all([x >= 0 for x in crop_roi]):
+        top, left, bottom, right = crop_roi
+        trans.append(transforms.Lambda(lambda x: TF.crop(x, top, left, bottom-top, right-left)))
     if resize[0] > 0 and resize[1] > 0:
-        return transforms.Compose([
-            transforms.Resize(resize),
-            transforms.ToTensor(),
-            transforms.Normalize(mean=[0.485, 0.456, 0.406],
-                                 std=[0.229, 0.224, 0.225]),
-        ])
-    else:
-        return transforms.Compose([
-            transforms.ToTensor(),
-            transforms.Normalize(mean=[0.485, 0.456, 0.406],
-                                 std=[0.229, 0.224, 0.225]),
-        ])
+        trans.append(transforms.Resize(resize))
+    trans.extend([
+        transforms.ToTensor(),
+        transforms.Normalize(mean=[0.485, 0.456, 0.406],
+                             std=[0.229, 0.224, 0.225]),
+    ])
+    return transforms.Compose(trans)
 
 
 class PlaceDataset(data.Dataset):
@@ -80,8 +80,11 @@ def __init__(self, query_file_path, index_file_path, dataset_root_dir, ground_tr
         self.positives = None
         self.distances = None
 
+        crop_roi = None
+        if 'imagecrop' in config:
+            crop_roi = tuple([int(x) for x in config['imagecrop'].split(',')])
         self.resize = (int(config['imageresizeH']), int(config['imageresizeW']))
-        self.mytransform = input_transform(self.resize)
+        self.mytransform = input_transform(self.resize, crop_roi)
 
     def __getitem__(self, index):
         img = Image.open(self.images[index])