Refine DensePose codes and fix minor bugs

INSTALL.md

@@ -14,7 +14,7 @@ The DensePose-RCNN system is implemented within the [`detectron`](https://github
 
 ## Caffe2
 
-To install Caffe2 with CUDA support, follow the [installation instructions](https://caffe2.ai/docs/getting-started.html) from the [Caffe2 website](https://caffe2.ai/). **If you already have Caffe2 installed, make sure to update your Caffe2 to a version that includes the [Detectron module](https://github.com/caffe2/caffe2/tree/master/modules/detectron).**
+To install Caffe2 with CUDA support, follow the [installation instructions](https://caffe2.ai/docs/getting-started.html) from the [Caffe2 website](https://caffe2.ai/). **If you already have Caffe2 installed, make sure to update your Caffe2 to a version that includes the [Detectron module](https://github.com/pytorch/pytorch/tree/master/modules/detectron).**
 
 Please ensure that your Caffe2 installation was successful before proceeding by running the following commands and checking their output as directed in the comments.
 
@@ -130,7 +130,7 @@ coco
 
 ## Docker Image
 
-We provide a [`Dockerfile`](docker/Dockerfile) that you can use to build a Densepose image on top of a Caffe2 image that satisfies the requirements outlined at the top. If you would like to use a Caffe2 image different from the one we use by default, please make sure that it includes the [Detectron module](https://github.com/caffe2/caffe2/tree/master/modules/detectron).
+We provide a [`Dockerfile`](docker/Dockerfile) that you can use to build a Densepose image on top of a Caffe2 image that satisfies the requirements outlined at the top. If you would like to use a Caffe2 image different from the one we use by default, please make sure that it includes the [Detectron module](https://github.com/pytorch/pytorch/tree/master/modules/detectron).
 
 Build the image:
 

README.md

@@ -6,7 +6,7 @@ _Rıza Alp Güler, Natalia Neverova, Iasonas Kokkinos_
 [[`densepose.org`](https://densepose.org)] [[`arXiv`](https://arxiv.org/abs/1802.00434)] [[`BibTeX`](#CitingDensePose)]
 
 Dense human pose estimation aims at mapping all human pixels of an RGB image to the 3D surface of the human body. 
-DensePose-RCNN is implemented in the [Detectron](https://github.com/facebookresearch/Detectron) framework and is powered by [Caffe2](https://github.com/caffe2/caffe2).
+DensePose-RCNN is implemented in the [Detectron](https://github.com/facebookresearch/Detectron) framework and is powered by [Caffe2](https://github.com/pytorch/pytorch/tree/master/caffe2).
 
 <div align="center">
   <img src="https://drive.google.com/uc?export=view&id=1qfSOkpueo1kVZbXOuQJJhyagKjMgepsz" width="700px" />

detectron/core/config.py

@@ -863,7 +863,7 @@
 __C.BODY_UV_RCNN.UP_SCALE = -1
 
 # Apply a ConvTranspose layer to the features prior to predicting the heatmaps
-__C.KRCNN.USE_DECONV = False
+__C.BODY_UV_RCNN.USE_DECONV = False
 # Channel dimension of the hidden representation produced by the ConvTranspose
 __C.BODY_UV_RCNN.DECONV_DIM = 256
 # Use a ConvTranspose layer to predict the heatmaps
@@ -876,6 +876,9 @@
 # Number of patches in the dataset
 __C.BODY_UV_RCNN.NUM_PATCHES = -1
 
+# Number of semantic parts used to sample annotation points
+__C.BODY_UV_RCNN.NUM_SEMANTIC_PARTS = 14
+
 # Number of stacked Conv layers in body UV head
 __C.BODY_UV_RCNN.NUM_STACKED_CONVS = 8
 # Dimension of the hidden representation output by the body UV head

detectron/core/test.py

@@ -948,16 +948,16 @@ def im_detect_body_uv(model, im_scale, boxes):
         # Removed squeeze calls due to singleton dimension issues
         CurAnnIndex = np.argmax(CurAnnIndex, axis=0)
         CurIndex_UV = np.argmax(CurIndex_UV, axis=0)
-        CurIndex_UV = CurIndex_UV * (CurAnnIndex>0).astype(np.float32)
+        CurIndex_UV = CurIndex_UV * (CurAnnIndex > 0).astype(np.float32)
 
         output = np.zeros([3, int(by), int(bx)], dtype=np.float32)
         output[0] = CurIndex_UV
 
         for part_id in range(1, K):
             CurrentU = CurU_uv[part_id]
             CurrentV = CurV_uv[part_id]
-            output[1, CurIndex_UV==part_id] = CurrentU[CurIndex_UV==part_id]
-            output[2, CurIndex_UV==part_id] = CurrentV[CurIndex_UV==part_id]
+            output[1, CurIndex_UV == part_id] = CurrentU[CurIndex_UV == part_id]
+            output[2, CurIndex_UV == part_id] = CurrentV[CurIndex_UV == part_id]
         outputs.append(output)
 
     num_classes = cfg.MODEL.NUM_CLASSES

detectron/core/test_engine.py

@@ -374,7 +374,7 @@ def get_roidb_and_dataset(dataset_name, proposal_file, ind_range):
 
 
 def empty_results(num_classes, num_images):
-    """Return empty results lists for boxes, masks, and keypoints.
+    """Return empty results lists for boxes, masks, keypoints and body IUVs.
     Box detections are collected into:
       all_boxes[cls][image] = N x 5 array with columns (x1, y1, x2, y2, score)
     Instance mask predictions are collected into:
@@ -386,8 +386,11 @@ def empty_results(num_classes, num_images):
       [x, y, logit, prob] (See: utils.keypoints.heatmaps_to_keypoints).
       Keypoints are recorded for person (cls = 1); they are in 1:1
       correspondence with the boxes in all_boxes[cls][image].
-    Body uv predictions are collected into:
-      TODO
+    Body IUV predictions are collected into:
+      all_bodys[cls][image] = [...] list of body IUV results that are in 1:1
+      correspondence with the boxes in all_boxes['person'][image], each encoded
+      as a 3D array (3, int(bbox_height), int(bbox_width)) with the 3 rows
+      corresponding to [Index, U, V].
     """
     # Note: do not be tempted to use [[] * N], which gives N references to the
     # *same* empty list.