Merge remote-tracking branch 'refs/remotes/thtrieu/master'

README.md

@@ -88,27 +88,27 @@ flow --h
 First, let's take a closer look at one of a very useful option `--load`
 
 ```bash
-# 1. Load yolo-tiny.weights
-flow --model cfg/yolo-tiny.cfg --load bin/yolo-tiny.weights
+# 1. Load tiny-yolo.weights
+flow --model cfg/tiny-yolo.cfg --load bin/tiny-yolo.weights
 
 # 2. To completely initialize a model, leave the --load option
 flow --model cfg/yolo-new.cfg
 
 # 3. It is useful to reuse the first identical layers of tiny for `yolo-new`
-flow --model cfg/yolo-new.cfg --load bin/yolo-tiny.weights
+flow --model cfg/yolo-new.cfg --load bin/tiny-yolo.weights
 # this will print out which layers are reused, which are initialized
 ```
 
 All input images from default folder `sample_img/` are flowed through the net and predictions are put in `sample_img/out/`. We can always specify more parameters for such forward passes, such as detection threshold, batch size, images folder, etc.
 
 ```bash
 # Forward all images in sample_img/ using tiny yolo and 100% GPU usage
-flow --imgdir sample_img/ --model cfg/yolo-tiny.cfg --load bin/yolo-tiny.weights --gpu 1.0
+flow --imgdir sample_img/ --model cfg/tiny-yolo.cfg --load bin/tiny-yolo.weights --gpu 1.0
 ```
 json output can be generated with descriptions of the pixel location of each bounding box and the pixel location. Each prediction is stored in the `sample_img/out` folder by default. An example json array is shown below.
 ```bash
 # Forward all images in sample_img/ using tiny yolo and JSON output.
-flow --imgdir sample_img/ --model cfg/yolo-tiny.cfg --load bin/yolo-tiny.weights --json
+flow --imgdir sample_img/ --model cfg/tiny-yolo.cfg --load bin/tiny-yolo.weights --json
 ```
 JSON output:
 ```json
@@ -127,7 +127,7 @@ Training is simple as you only have to add option `--train`. Training set and an
 
 ```bash
 # Initialize yolo-new from yolo-tiny, then train the net on 100% GPU:
-flow --model cfg/yolo-new.cfg --load bin/yolo-tiny.weights --train --gpu 1.0
+flow --model cfg/yolo-new.cfg --load bin/tiny-yolo.weights --train --gpu 1.0
 
 # Completely initialize yolo-new and train it with ADAM optimizer
 flow --model cfg/yolo-new.cfg --train --trainer adam
@@ -143,7 +143,7 @@ flow --train --model cfg/yolo-new.cfg --load -1
 flow --model cfg/yolo-new.cfg --load 1500
 
 # Fine tuning yolo-tiny from the original one
-flow --train --model cfg/yolo-tiny.cfg --load bin/yolo-tiny.weights
+flow --train --model cfg/tiny-yolo.cfg --load bin/tiny-yolo.weights
 ```
 
 Example of training on Pascal VOC 2007:
@@ -248,7 +248,7 @@ options = {"model": "cfg/yolo.cfg", "load": "bin/yolo.weights", "threshold": 0.1
 
 tfnet = TFNet(options)
 
-imgcv = cv2.imread("./sample_img/dog.jpg")
+imgcv = cv2.imread("./sample_img/sample_dog.jpg")
 result = tfnet.return_predict(imgcv)
 print(result)
 ```

darkflow/net/flow.py

@@ -120,15 +120,10 @@ def predict(self):
         to_idx = min(from_idx + batch, len(all_inps))
 
         # collect images input in the batch
-        inp_feed = list(); new_all = list()
         this_batch = all_inps[from_idx:to_idx]
-        for inp in this_batch:
-            new_all += [inp]
-            this_inp = os.path.join(inp_path, inp)
-            this_inp = self.framework.preprocess(this_inp)
-            expanded = np.expand_dims(this_inp, 0)
-            inp_feed.append(expanded)
-        this_batch = new_all
+        inp_feed = pool.map(lambda inp: (
+            np.expand_dims(self.framework.preprocess(
+                os.path.join(inp_path, inp)), 0)), this_batch)
 
         # Feed to the net
         feed_dict = {self.inp : np.concatenate(inp_feed, 0)}    

darkflow/net/help.py

@@ -20,7 +20,7 @@ def build_train_op(self):
 
 def load_from_ckpt(self):
     if self.FLAGS.load < 0: # load lastest ckpt
-        with open(self.FLAGS.backup + 'checkpoint', 'r') as f:
+        with open(os.path.join(self.FLAGS.backup, 'checkpoint'), 'r') as f:
             last = f.readlines()[-1].strip()
             load_point = last.split(' ')[1]
             load_point = load_point.split('"')[1]

darkflow/net/yolo/misc.py

@@ -43,7 +43,7 @@ def labels(meta, FLAGS):
         meta['labels'] = labels20
 
 def is_inp(self, name): 
-    return name[-4:] in ['.jpg','.JPG', '.jpeg', '.JPEG', '.png', '.PNG']
+    return name.lower().endswith(('.jpg', '.jpeg', '.png'))
 
 def show(im, allobj, S, w, h, cellx, celly):
     for obj in allobj:
@@ -59,7 +59,7 @@ def show(im, allobj, S, w, h, cellx, celly):
             (int(centerx - ww/2), int(centery - hh/2)),
             (int(centerx + ww/2), int(centery + hh/2)),
             (0,0,255), 2)
-    cv2.imshow("result", im)
+    cv2.imshow('result', im)
     cv2.waitKey()
     cv2.destroyAllWindows()
 

darkflow/utils/im_transform.py

@@ -11,7 +11,8 @@ def imcv2_recolor(im, a = .1):
 	im = im * (1 + t * a)
 	mx = 255. * (1 + a)
 	up = np.random.uniform() * 2 - 1
-	im = np.power(im/mx, 1. + up * .5)
+# 	im = np.power(im/mx, 1. + up * .5)
+	im = cv2.pow(im/mx, 1. + up * .5)
 	return np.array(im * 255., np.uint8)
 
 def imcv2_affine_trans(im):
@@ -27,4 +28,4 @@ def imcv2_affine_trans(im):
 	im = im[offy : (offy + h), offx : (offx + w)]
 	flip = np.random.binomial(1, .5)
 	if flip: im = cv2.flip(im, 1)
-	return im, [w, h, c], [scale, [offx, offy], flip]
+	return im, [w, h, c], [scale, [offx, offy], flip]