update fpn

examples/FasterRCNN/README.md

 # Faster-RCNN / Mask-RCNN on COCO
 This example aims to provide a minimal (1.3k lines) implementation of
-end-to-end Faster-RCNN & Mask-RCNN (with ResNet backbones) on COCO.
+end-to-end Faster-RCNN & Mask-RCNN (with ResNet & FPN backbones) on COCO.
 
 ## Dependencies
-+ Python 3; TensorFlow >= 1.4.0
++ Python 3; TensorFlow >= 1.4.0 (>=1.6.0 recommended due to a TF bug);
 + [pycocotools](https://github.com/pdollar/coco/tree/master/PythonAPI/pycocotools), OpenCV.
 + Pre-trained [ResNet model](http://models.tensorpack.com/ResNet/) from tensorpack model zoo.
 + COCO data. It assumes the following directory structure:
@@ -61,7 +61,7 @@ MaskRCNN results contain both bbox and segm mAP.
 |R-101   |512             |(800, 1333)|280k    |40.1/34.4      |70h on 8 P100s|
 |R-101   |512             |(800, 1333)|360k    |40.8/35.1      |63h on 8 V100s|
 
-The two R-50 360k models have the same configuration __and mAP__ 
+The two R-50 360k models have the same configuration __and mAP__
 as the `R50-C4-2x` entries in
 [Detectron Model Zoo](https://github.com/facebookresearch/Detectron/blob/master/MODEL_ZOO.md#end-to-end-faster--mask-r-cnn-baselines).
 So far this seems to be the only open source re-implementation that can reproduce mAP in Detectron.

examples/FasterRCNN/config.py

@@ -5,11 +5,11 @@ import numpy as np
 
 # mode flags ---------------------
 MODE_MASK = True
+MODE_FPN = False
 
 # dataset -----------------------
 BASEDIR = '/path/to/your/COCO/DIR'
 TRAIN_DATASET = ['train2014', 'valminusminival2014']
-# TRAIN_DATASET = ['valminusminival2014']
 VAL_DATASET = 'minival2014'   # only support evaluation on single dataset
 NUM_CLASS = 81
 CLASS_NAMES = []  # NUM_CLASS strings. Will be populated later by coco loader
@@ -29,14 +29,14 @@ LR_SCHEDULE = [240000, 320000, 360000]    # "2x" schedule in detectron
 
 # image resolution --------------------
 SHORT_EDGE_SIZE = 800
-MAX_SIZE = 1333  # TODO use 1344
+MAX_SIZE = 1333
 # alternative (worse & faster) setting: 600, 1024
 
 # anchors -------------------------
 ANCHOR_STRIDE = 16
 ANCHOR_STRIDES_FPN = (4, 8, 16, 32, 64)
-# sqrtarea of the anchor box
-ANCHOR_SIZES = (32, 64, 128, 256, 512)
+FPN_RESOLUTION_REQUIREMENT = 32    # image size into the backbone has to be multiple of this number
+ANCHOR_SIZES = (32, 64, 128, 256, 512)   # sqrtarea of the anchor box
 ANCHOR_RATIOS = (0.5, 1., 2.)
 NUM_ANCHOR = len(ANCHOR_SIZES) * len(ANCHOR_RATIOS)
 POSITIVE_ANCHOR_THRES = 0.7
@@ -52,6 +52,7 @@ RPN_MIN_SIZE = 0
 RPN_PROPOSAL_NMS_THRESH = 0.7
 TRAIN_PRE_NMS_TOPK = 12000
 TRAIN_POST_NMS_TOPK = 2000
+TRAIN_FPN_NMS_TOPK = 2000
 # boxes overlapping crowd will be ignored.
 CROWD_OVERLAP_THRES = 0.7
 
@@ -62,19 +63,16 @@ FASTRCNN_FG_THRESH = 0.5
 # fg ratio in a ROI batch
 FASTRCNN_FG_RATIO = 0.25
 
+# modeling -------------------------
+FPN_NUM_CHANNEL = 256
+FASTRCNN_FC_HEAD_DIM = 1024
+MASKRCNN_HEAD_DIM = 256
+
 # testing -----------------------
 TEST_PRE_NMS_TOPK = 6000
 TEST_POST_NMS_TOPK = 1000   # if you encounter OOM in inference, set this to a smaller number
+TEST_FPN_NMS_TOPK = 1000
 FASTRCNN_NMS_THRESH = 0.5
 RESULT_SCORE_THRESH = 0.05
 RESULT_SCORE_THRESH_VIS = 0.3   # only visualize confident results
 RESULTS_PER_IM = 100
-
-# TODO Not Functioning. Don't USE
-MODE_FPN = True
-FPN_NUM_CHANNEL = 256
-MASKRCNN_HEAD_DIM = 256
-FASTRCNN_FC_HEAD_DIM = 1024
-FPN_RESOLUTION_REQUIREMENT = 32
-TRAIN_FPN_NMS_TOPK = 2000
-TEST_FPN_NMS_TOPK = 1000

examples/FasterRCNN/data.py

@@ -8,7 +8,7 @@ import itertools
 
 from tensorpack.utils.argtools import memoized, log_once
 from tensorpack.dataflow import (
-    imgaug, TestDataSpeed, PrefetchDataZMQ, MapData, MultiProcessMapDataZMQ,
+    imgaug, TestDataSpeed, PrefetchDataZMQ, MultiProcessMapDataZMQ,
     MapDataComponent, DataFromList)
 # import tensorpack.utils.viz as tpviz
 

examples/FasterRCNN/model.py

@@ -12,7 +12,6 @@ from tensorpack.models import (
     Conv2D, FullyConnected, MaxPooling,
     layer_register, Conv2DTranspose, FixedUnPooling)
 
-from tensorpack.utils import logger
 from utils.box_ops import pairwise_iou
 from utils.box_ops import area as tf_area
 import config
@@ -90,7 +89,7 @@ def rpn_losses(anchor_labels, anchor_boxes, label_logits, box_logits):
                         valid_label_prob > th,
                         tf.equal(valid_prediction, valid_anchor_labels)),
                     dtype=tf.int32)
-                placeholder = 0.5   # TODO A small value will make summaries appear lower.
+                placeholder = 0.5   # A small value will make summaries appear lower.
                 recall = tf.to_float(tf.truediv(pos_prediction_corr, nr_pos))
                 recall = tf.where(tf.equal(nr_pos, 0), placeholder, recall, name='recall_th{}'.format(th))
                 precision = tf.to_float(tf.truediv(pos_prediction_corr, nr_pos_prediction))
@@ -99,7 +98,9 @@ def rpn_losses(anchor_labels, anchor_boxes, label_logits, box_logits):
                 summaries.extend([precision, recall])
         add_moving_summary(*summaries)
 
-    placeholder = 0.    # Per-level loss summaries in FPN may appear lower. But the sum should be OK.
+    # Per-level loss summaries in FPN may appear lower due to the use of a small placeholder.
+    # But the total loss is still the same.
+    placeholder = 0.
     label_loss = tf.nn.sigmoid_cross_entropy_with_logits(
         labels=tf.to_float(valid_anchor_labels), logits=valid_label_logits)
     label_loss = tf.reduce_sum(label_loss) * (1. / config.RPN_BATCH_PER_IM)
@@ -601,19 +602,18 @@ def fpn_model(features):
     num_channel = config.FPN_NUM_CHANNEL
 
     def upsample2x(name, x):
-        # TODO may not be optimal in speed or math
-        logger.info("Unpool 1111 ...")
         return FixedUnPooling(
             name, x, 2, unpool_mat=np.ones((2, 2), dtype='float32'),
             data_format='channels_first')
 
-        with tf.name_scope(name):
-            logger.info("Nearest neighbor")
-            shape2d = tf.shape(x)[2:]
-            x = tf.transpose(x, [0, 2, 3, 1])
-            x = tf.image.resize_nearest_neighbor(x, shape2d * 2, align_corners=True)
-            x = tf.transpose(x, [0, 3, 1, 2])
-            return x
+        # tf.image.resize is, again, not aligned.
+        # with tf.name_scope(name):
+        #     logger.info("Nearest neighbor")
+        #     shape2d = tf.shape(x)[2:]
+        #     x = tf.transpose(x, [0, 2, 3, 1])
+        #     x = tf.image.resize_nearest_neighbor(x, shape2d * 2, align_corners=True)
+        #     x = tf.transpose(x, [0, 3, 1, 2])
+        #     return x
 
     with argscope(Conv2D, data_format='channels_first',
                   nl=tf.identity, use_bias=True,
@@ -636,6 +636,8 @@ def fpn_model(features):
 @under_name_scope()
 def fpn_map_rois_to_levels(boxes):
     """
+    Assign boxes to level 2~5.
+
     Args:
         boxes (nx4)
 

examples/FasterRCNN/train.py

@@ -19,6 +19,7 @@ from tensorpack import *
 from tensorpack.tfutils.summary import add_moving_summary
 from tensorpack.tfutils.scope_utils import under_name_scope
 from tensorpack.tfutils import optimizer
+from tensorpack.tfutils.common import get_tf_version_number
 import tensorpack.utils.viz as tpviz
 from tensorpack.utils.gpu import get_nr_gpu
 
@@ -33,8 +34,7 @@ from model import (
     generate_rpn_proposals, sample_fast_rcnn_targets, roi_align,
     fastrcnn_outputs, fastrcnn_losses, fastrcnn_predictions,
     maskrcnn_upXconv_head, maskrcnn_loss,
-    fpn_model, fpn_map_rois_to_levels, fastrcnn_2fc_head,
-    multilevel_roi_align)
+    fpn_model, fastrcnn_2fc_head, multilevel_roi_align)
 from data import (
     get_train_dataflow, get_eval_dataflow,
     get_all_anchors, get_all_anchors_fpn)
@@ -62,6 +62,8 @@ def get_model_output_names():
 
 def get_model():
     if config.MODE_FPN:
+        if get_tf_version() < 1.6:
+            logger.warn("FPN has chances to crash in TF<1.6, due to a TF issue.")
         return ResNetFPNModel()
     else:
         return ResNetC4Model()
@@ -223,8 +225,12 @@ class ResNetC4Model(DetectionModel):
             ncls = config.NUM_CLASS
             return tf.zeros([0, 2048, 7, 7]), tf.zeros([0, ncls]), tf.zeros([0, ncls - 1, 4])
 
-        feature_fastrcnn, fastrcnn_label_logits, fastrcnn_box_logits = tf.cond(
-            tf.size(boxes_on_featuremap) > 0, ff_true, ff_false)
+        if get_tf_version_number() >= 1.6:
+            feature_fastrcnn, fastrcnn_label_logits, fastrcnn_box_logits = ff_true()
+        else:
+            logger.warn("This example may drop support for TF < 1.6 soon.")
+            feature_fastrcnn, fastrcnn_label_logits, fastrcnn_box_logits = tf.cond(
+                tf.size(boxes_on_featuremap) > 0, ff_true, ff_false)
 
         if is_training:
             # rpn loss
@@ -434,10 +440,11 @@ class ResNetFPNModel(DetectionModel):
                     'maskrcnn', roi_feature_maskrcnn, config.NUM_CLASS, 4)   # #fg x #cat x 28 x 28
                 indices = tf.stack([tf.range(tf.size(final_labels)), tf.to_int32(final_labels) - 1], axis=1)
                 final_mask_logits = tf.gather_nd(mask_logits, indices)   # #resultx28x28
-                final_masks = tf.sigmoid(final_mask_logits, name='final_masks')
+                tf.sigmoid(final_mask_logits, name='final_masks')
 
 
 def visualize(model_path, nr_visualize=50, output_dir='output'):
+    assert not config.MODE_FPN, "FPN visualize is not supported yet!"
     df = get_train_dataflow()   # we don't visualize mask stuff
     df.reset_state()
 
@@ -577,7 +584,7 @@ if __name__ == '__main__':
                 COCODetection(config.BASEDIR, 'val2014')   # Only to load the class names into caches
                 predict(pred, args.predict)
     else:
-        logger.set_logger_dir(args.logdir, 'd')
+        logger.set_logger_dir(args.logdir)
         print_config()
         factor = get_batch_factor()
         stepnum = config.STEPS_PER_EPOCH
@@ -611,5 +618,5 @@ if __name__ == '__main__':
             max_epoch=config.LR_SCHEDULE[-1] * factor // stepnum,
             session_init=get_model_loader(args.load) if args.load else None,
         )
-        trainer = SyncMultiGPUTrainerReplicated(get_nr_gpu())
+        trainer = SyncMultiGPUTrainerReplicated(get_nr_gpu(), mode='cpu')
         launch_train_with_config(cfg, trainer)