re-eval: option to use mat files; some pylint fixes

lib/fast_rcnn/config.py

@@ -120,7 +120,7 @@ __C.RNG_SEED = 3
 __C.EPS = 1e-14
 
 # Root directory of project
-__C.ROOT_DIR = ops.abspath(osp.join(osp.dirname(__file__), '..', '..'))
+__C.ROOT_DIR = osp.abspath(osp.join(osp.dirname(__file__), '..', '..'))
 
 # Place outputs under an experiments directory
 __C.EXP_DIR = 'default'

lib/fast_rcnn/test.py

@@ -14,11 +14,10 @@ import caffe
 import utils.cython_nms
 import cPickle
 import heapq
-import utils.blob
+from utils.blob import im_list_to_blob
 import os
 
 def _get_image_blob(im):
-    im_pyra = []
     im_orig = im.astype(np.float32, copy=True)
     im_orig -= cfg.PIXEL_MEANS
 
@@ -40,7 +39,7 @@ def _get_image_blob(im):
         processed_ims.append(im)
 
     # Create a blob to hold the input images
-    blob = utils.blob.im_list_to_blob(processed_ims)
+    blob = im_list_to_blob(processed_ims)
 
     return blob, np.array(im_scale_factors)
 
@@ -117,7 +116,7 @@ def _clip_boxes(boxes, im_shape):
     return boxes
 
 def im_detect(net, im, boxes):
-    blobs, im_scale_factors = _get_blobs(im, boxes)
+    blobs, unused_im_scale_factors = _get_blobs(im, boxes)
 
     # When mapping from image ROIs to feature map ROIs, there's some aliasing
     # (some distinct image ROIs get mapped to the same feature ROI).

lib/fast_rcnn/train.py

@@ -7,10 +7,9 @@
 
 from fast_rcnn.config import cfg, get_output_path
 import numpy as np
-import cv2
 import caffe
-import finetuning
-import bbox_regression_targets
+import fast_rcnn.finetuning as finetuning
+import fast_rcnn.bbox_regression_targets as bbox_regression_targets
 import os
 
 from caffe.proto import caffe_pb2
@@ -77,7 +76,7 @@ class SolverWrapper(object):
             for shuffled_i in xrange(0, len(shuffled_inds),
                                      cfg.TRAIN.IMS_PER_BATCH):
                 db_inds = shuffled_inds[shuffled_i:shuffled_i +
-                            cfg.TRAIN.IMS_PER_BATCH]
+                                        cfg.TRAIN.IMS_PER_BATCH]
                 minibatch_db = [roidb[i] for i in db_inds]
                 blobs = finetuning.get_minibatch(minibatch_db)
 

tools/reval.py

@@ -6,8 +6,29 @@ from fast_rcnn.config import cfg
 from datasets.factory import get_imdb
 import cPickle
 import os, sys
+import numpy as np
 
-def main(imdb_name, output_dir):
+def from_mats(imdb_name, output_dir):
+    import scipy.io as sio
+
+    imdb = get_imdb(imdb_name)
+    aps = []
+    for i, cls in enumerate(imdb.classes[1:]):
+        mat = sio.loadmat(os.path.join(output_dir, cls + '_pr.mat'))
+        ap = mat['ap'][0, 0] * 100
+        apAuC = mat['ap_auc'][0, 0] * 100
+        print '!!! {} : {:.1f} {:.1f}'.format(cls, ap, apAuC)
+        aps.append(ap)
+
+    print '~~~~~~~~~~~~~~~~~~~'
+    print 'Results (from mat files):'
+    for ap in aps:
+        print '{:.1f}'.format(ap)
+    print '{:.1f}'.format(np.array(aps).mean())
+    print '~~~~~~~~~~~~~~~~~~~'
+
+
+def from_dets(imdb_name, output_dir):
     imdb = get_imdb(imdb_name)
     imdb.config['use_salt'] = False
     imdb.config['cleanup'] = False
@@ -26,4 +47,8 @@ if __name__ == '__main__':
     output_dir = os.path.abspath(os.path.join(os.path.dirname(__file__),
                                  '..', output_dir))
     imdb_name = 'voc_2007_test'
-    main(imdb_name, output_dir)
+
+    if len(sys.argv) > 2:
+        from_mats(imdb_name, output_dir)
+    else:
+        from_dets(imdb_name, output_dir)