change where nets and results are output; add EXP_DIR to config

src/fast_rcnn_config.py

@@ -122,7 +122,17 @@ __C.EPS             = 1e-14
 # Root directory of project
 __C.ROOT_DIR        = osp.abspath(osp.join(osp.dirname(__file__), '..'))
 
-def merge_a_into_b(a, b):
+# Place outputs under an experiments directory
+__C.EXP_DIR         = 'default'
+
+def get_output_path(imdb, net):
+    path = os.path.join(__C.ROOT_DIR, 'output', __C.EXP_DIR, imdb.name)
+    if net is None:
+        return path
+    else:
+        return os.path.join(path, net.name)
+
+def _merge_a_into_b(a, b):
     """
     Merge config dictionary a into config dictionary b, clobbering the options
     in b whenever they are also specified in a.
@@ -142,7 +152,7 @@ def merge_a_into_b(a, b):
         # recursively merge dicts
         if type(v) is edict:
             try:
-                merge_a_into_b(a[k], b[k])
+                _merge_a_into_b(a[k], b[k])
             except:
                 print('Error under config key: {}'.format(k))
                 raise
@@ -159,4 +169,4 @@ def cfg_from_file(filename):
     with open(filename, 'r') as f:
         yaml_cfg = edict(yaml.load(f))
 
-    merge_a_into_b(yaml_cfg, __C)
+    _merge_a_into_b(yaml_cfg, __C)

src/fast_rcnn_test.py

@@ -5,7 +5,7 @@
 # Written by Ross Girshick
 # --------------------------------------------------------
 
-from fast_rcnn_config import cfg
+from fast_rcnn_config import cfg, get_output_path
 import argparse
 from utils.timer import Timer
 import numpy as np
@@ -213,9 +213,7 @@ def test_net(net, imdb):
     all_boxes = [[[] for _ in xrange(num_images)]
                  for _ in xrange(imdb.num_classes)]
 
-    # Output directory will be something like:
-    #   output/vgg16_fast_rcnn_iter_40000/voc_2007_test/
-    output_dir = os.path.join(cfg.ROOT_DIR, 'output', net.name, imdb.name)
+    output_dir = get_output_path(imdb, net)
     if not os.path.exists(output_dir):
         os.makedirs(output_dir)
 

src/fast_rcnn_train.py

@@ -5,20 +5,22 @@
 # Written by Ross Girshick
 # --------------------------------------------------------
 
-from fast_rcnn_config import cfg
+from fast_rcnn_config import cfg, get_output_path
 import numpy as np
 import cv2
 import caffe
 import finetuning
 import bbox_regression_targets
+import os
 
 from caffe.proto import caffe_pb2
 import google.protobuf as pb2
 
 class SolverWrapper(object):
-    def __init__(self, solver_prototxt, pretrained_model=None):
+    def __init__(self, solver_prototxt, imdb, pretrained_model=None):
         self.bbox_means = None
         self.bbox_stds = None
+        self.imdb = imdb
 
         self.solver = caffe.SGDSolver(solver_prototxt)
         if pretrained_model is not None:
@@ -47,10 +49,16 @@ class SolverWrapper(object):
         self.solver.net.params['bbox_pred'][1].data[...] = \
                 self.solver.net.params['bbox_pred'][1].data + means
 
+        output_dir = get_output_path(self.imdb, None)
+        if not os.path.exists(output_dir):
+            os.makedirs(output_dir)
+
         infix = ('_' + cfg.TRAIN.SNAPSHOT_INFIX
                  if cfg.TRAIN.SNAPSHOT_INFIX != '' else '')
         filename = self.solver_param.snapshot_prefix + infix + \
-              '_iter_{:d}'.format(self.solver.iter) + '.caffemodel'
+                   '_iter_{:d}'.format(self.solver.iter) + '.caffemodel'
+        filename = os.path.join(output_dir, filename)
+
         self.solver.net.save(str(filename))
         print 'Wrote snapshot to: {:s}'.format(filename)
 
@@ -169,7 +177,7 @@ def train_net(solver_prototxt, imdb, pretrained_model=None, max_iters=40000):
         bbox_regression_targets.append_bbox_regression_targets(roidb)
     print 'done'
 
-    sw = SolverWrapper(solver_prototxt, pretrained_model=pretrained_model)
+    sw = SolverWrapper(solver_prototxt, imdb, pretrained_model=pretrained_model)
     sw.bbox_means = means
     sw.bbox_stds = stds