MULTICLASS NMS. DATA AUG.

README.md

@@ -13,15 +13,22 @@ Original paper and github: [YOLO9000: Better, Faster, Stronger](https://arxiv.or
 ### Dependancies
 ---
 - [keras](https://github.com/fchollet/keras)
-- [tensorflow](https://www.tensorflow.org/)
+- [tensorflow 1.9](https://www.tensorflow.org/api_docs/)
 - [numpy](http://www.numpy.org/)
 - [h5py](http://www.h5py.org/)
 - [opencv](https://pypi.org/project/opencv-python/)
-- [python3.5](https://www.python.org/)
+- [python 3.5](https://www.python.org/)
 - [moviepy](https://zulko.github.io/moviepy/) (optional, gifs)
 ### Simple use
 ---
+
+0. Clone repository: 
+```bash
+git clone https://github.com/ksanjeevan/dourflow.git
+```
+
 1. Download pretrained [model](https://drive.google.com/open?id=1khOgS8VD-paUD8KhjOLOzEXkzaXNAAMq) and place it in **dourflow/**.
+
 2. Predict on an [image](https://images.pexels.com/photos/349758/hummingbird-bird-birds-349758.jpeg?auto=compress&cs=tinysrgb&h=350):
 
 ```bash
@@ -38,13 +45,14 @@ Running `python3 dourflow.py --help`:
 
 ```bash
 usage: dourflow.py [-h] [-m MODEL] [-c CONF] [-t THRESHOLD] [-w WEIGHT_FILE]
+                   [--gif]
                    action
 
 dourflow: a keras YOLO V2 implementation.
 
 positional arguments:
-  action                what to do: 'train', 'validate', 'cam' or pass a video, image
-                        file/dir.
+  action                what to do: 'train', 'validate', 'cam' or pass a
+                        video, image file/dir.
 
 optional arguments:
   -h, --help            show this help message and exit
@@ -55,7 +63,7 @@ optional arguments:
                         detection threshold
   -w WEIGHT_FILE, --weight_file WEIGHT_FILE
                         path to weight file
-
+  --gif                 video output stored as gif also
 
 ```
 #### *action* [positional]
@@ -151,28 +159,28 @@ Output:
 
 ```bash
 Batch Processed: 100%|████████████████████████████████████████████| 4282/4282 [01:53<00:00, 37.84it/s]
-AP( bus ): 0.806
-AP( tvmonitor ): 0.716
-AP( motorbike ): 0.666
-AP( dog ): 0.811
-AP( horse ): 0.574
-AP( boat ): 0.618
-AP( sofa ): 0.625
-AP( sheep ): 0.718
-AP( bicycle ): 0.557
-AP( cow ): 0.725
-AP( pottedplant ): 0.565
-AP( train ): 0.907
-AP( bird ): 0.813
-AP( person ): 0.665
-AP( car ): 0.580
 AP( cat ): 0.908
-AP( bottle ): 0.429
-AP( diningtable ): 0.593
-AP( chair ): 0.475
-AP( aeroplane ): 0.724
+AP( train ): 0.907
+AP( dog ): 0.899
+AP( bird ): 0.814
+AP( aeroplane ): 0.810
+AP( cow ): 0.810
+AP( bus ): 0.806
+AP( motorbike ): 0.792
+AP( person ): 0.737
+AP( sheep ): 0.719
+AP( tvmonitor ): 0.718
+AP( sofa ): 0.701
+AP( bicycle ): 0.683
+AP( diningtable ): 0.665
+AP( car ): 0.641
+AP( boat ): 0.617
+AP( horse ): 0.575
+AP( pottedplant ): 0.568
+AP( chair ): 0.528
+AP( bottle ): 0.487
 -------------------------------
-mAP: 0.674
+mAP: 0.719
 
 ```
 
@@ -198,7 +206,7 @@ Will store the custom bounding box priors wherever the path indicates in the con
 
 Training will create directory **logs/** which will store metrics and checkpoints for all the different training runs.
 
-Model passed is used for [transfer learning](https://en.wikipedia.org/wiki/Transfer_learning) (TRAINING FROM SCRATCH / TRAINING ONLY LAST LAYER SHOULD BE ADDED SOON).
+Model passed is used for [transfer learning](https://en.wikipedia.org/wiki/Transfer_learning).
 
 Example:
 `python3 dourflow.py train -m models/logo/coco_model.h5 -c confs/config_custom.json`
@@ -214,16 +222,16 @@ Then, in another terminal tab you can run `tensorboard --logdir=logs/run_X` and
 
 #### To Do
 
-- [ ] cfg parser
+- [x] Multiclass NMS
 - [x] Anchor generation for custom datasets
 - [ ] mAP write up
 - [x] Add webcam support
-- [ ] Data Augmentation
+- [x] Data Augmentation
 - [x] TensorBoard metrics
 
 #### Inspired from
 
 - [Darknet](https://github.com/pjreddie/darknet)
 - [Darkflow](https://github.com/thtrieu/darkflow)
 - [keras-yolo2](https://github.com/experiencor/keras-yolo2)
-- [YAD2K](https://github.com/allanzelener/YAD2K)
+- [YAD2K](https://github.com/allanzelener/YAD2K)

confs/config_coco.json

@@ -2,9 +2,9 @@
     "model" : {
         "input_size":       416,
         "grid_size":        13,
-        "true_box_buffer":  30,
+        "true_box_buffer":  10,
         "iou_threshold":    0.5,  
-        "nms_threshold":    0.3
+        "nms_threshold":    0.4
     },
     "config_path" : {
         "labels":           "models/coco/labels_coco.txt",

confs/config_voc.json

@@ -4,7 +4,7 @@
         "grid_size":        13,
         "true_box_buffer":  10,
         "iou_threshold":    0.5,  
-        "nms_threshold":    0.3
+        "nms_threshold":    0.45
     },
     "config_path" : {
         "labels":           "models/voc/labels_voc.txt",
@@ -17,7 +17,7 @@
         "annot_folder":     "/home/kiran/Documents/DATA/VOC/train/anns",
         "batch_size":       16,
         "learning_rate":    1e-4,
-        "num_epochs":       20,
+        "num_epochs":       50,
         "object_scale":     5.0 ,
         "no_object_scale":  1.0,
         "coord_scale":      1.0,

kmeans_anchors.py

@@ -105,7 +105,6 @@ def IoU_dist(x, c):
 
 
 
-
 def exrtract_wh(img):
     result = []
     pixel_height = img['height']

net/__init__.py

@@ -0,0 +1,7 @@
+#from . import netarch
+#from . import netdecode
+#from . import neteval
+#from . import netgen
+#from . import netloss
+#from . import netparams
+#from . import utils

net/netarch.py

@@ -1,17 +1,20 @@
+"""
+Set up keras model with Yolo v2 architecture, for both training
+and inference.
+"""
+import tensorflow as tf
+import numpy as np
+import pickle, argparse, json, os, cv2
+
 from keras.models import Model, load_model
 from keras.layers import Reshape, Conv2D, Input, MaxPooling2D, BatchNormalization, Lambda
 from keras.layers.advanced_activations import LeakyReLU
 
 from keras.layers.merge import concatenate
-
-import tensorflow as tf
-import numpy as np
-import pickle, argparse, json, os, cv2
-
 from keras.utils.vis_utils import plot_model
 
-from net.netparams import YoloParams
-from net.netdecode import YoloOutProcess
+from .netparams import YoloParams
+from .netdecode import YoloOutProcess
 
 
 
@@ -20,6 +23,7 @@ class YoloInferenceModel(object):
     def __init__(self, model):
         self._yolo_out = YoloOutProcess()
         self._inf_model = self._extend_processing(model)
+        self._model = model
 
     def _extend_processing(self, model):
         output = Lambda(self._yolo_out, name='lambda_2')(model.output)
@@ -40,6 +44,8 @@ def _prepro_single_image(self, image):
     def predict(self, image):
 
         image = self._prepro_single_image(image)
+
+        #np.save('person.npy', self._model.predict(image))
         output = self._inf_model.predict(image)[0]
 
         if output.size == 0: