Update train instructions.

donhk1011 · Jun 7, 2019 · f31da5f · f31da5f
1 parent 250da28
commit f31da5f
Show file tree

Hide file tree

Showing 3 changed files with 23 additions and 15 deletions.
diff --git a/README.md b/README.md
@@ -124,7 +124,7 @@ git clone https://github.com/SamvitJ/Accel.git
 ## Demo
 
 1. To run the demo with our trained models, please download the following models, and place them under folder `model/`:
-	- FlowNet model -- manually from [OneDrive](https://1drv.ms/u/s!Am-5JzdW2XHzhqMPLjGGCvAeciQflg) (for users from Mainland China, please try [Baidu Yun](https://pan.baidu.com/s/1nuPULnj))
+	- Base DFF model (with FlowNet) -- manually from [OneDrive](https://1drv.ms/u/s!Am-5JzdW2XHzhqMPLjGGCvAeciQflg) (for users in Mainland China, please try [Baidu Yun](https://pan.baidu.com/s/1nuPULnj))
 	- Accel models -- manually from [Google Drive]() [TODO: ADD LINK]
 
 	Make sure it looks like this:
@@ -145,23 +145,31 @@ git clone https://github.com/SamvitJ/Accel.git
 	```
 
 
-## Preparation for Training & Testing [UPDATE]
+## Preparation for Training & Testing
 
-1. Please download ILSVRC2015 DET and ILSVRC2015 VID dataset, and make sure it looks like this:
+1. Please download the Cityscapes dataset, and make sure the directories look like this:
 
 	```
-	./data/ILSVRC2015/
-	./data/ILSVRC2015/Annotations/DET
-	./data/ILSVRC2015/Annotations/VID
-	./data/ILSVRC2015/Data/DET
-	./data/ILSVRC2015/Data/VID
-	./data/ILSVRC2015/ImageSets
+	./data/cityscapes
+	./data/cityscapes/leftImg8bit_sequence/train
+	./data/cityscapes/leftImg8bit_sequence/val
+	./data/cityscapes/leftImg8bit_sequence/test
+	./data/cityscapes/gtFine/train
+	./data/cityscapes/gtFine/val
+	./data/cityscapes/gtFine/test
 	```
 
-2. Please download ImageNet pre-trained ResNet-v1-101 model and Flying-Chairs pre-trained FlowNet model manually from [OneDrive](https://1drv.ms/u/s!Am-5JzdW2XHzhqMOBdCBiNaKbcjPrA) (for users from Mainland China, please try [Baidu Yun](https://pan.baidu.com/s/1nuPULnj)), and put it under folder `./model`. Make sure it looks like this:
+2. Please download the following models, and place them under folder `./model`:
+	- Base DFF model (with FlowNet) -- manually from [OneDrive](https://1drv.ms/u/s!Am-5JzdW2XHzhqMOBdCBiNaKbcjPrA) (for users in Mainland China, please try [Baidu Yun](https://pan.baidu.com/s/1nuPULnj))
+	- DeepLab models -- manually from [Google Drive]() [TODO: ADD LINK]
+
+	Make sure it looks like this:
 	```
-	./model/pretrained_model/resnet_v1_101-0000.params
-	./model/pretrained_model/flownet-0000.params
+	./model/rfcn_dff_flownet_vid-0000.params
+	./model/pretrained/deeplab-18-0000.params
+	./model/pretrained/deeplab-34-0000.params
+	./model/pretrained/deeplab-50-0000.params
+	./model/pretrained/deeplab-101-0000.params
 	```
 
 

diff --git a/experiments/dff_deeplab/cfgs/resnet_v1_101_flownet_cityscapes_deeplab_end2end_ohem.yaml b/experiments/dff_deeplab/cfgs/resnet_v1_101_flownet_cityscapes_deeplab_end2end_ohem.yaml
@@ -11,8 +11,8 @@ default:
   kvstore: device
 network:
   pretrained: "./model/rfcn_dff_flownet_vid"
-  pretrained_flow: "./model/deeplab_dcn_cityscapes"
-  pretrained_ec: "./model/trained/resnet-18*-0092"
+  pretrained_flow: "./model/pretrained/deeplab-101"
+  pretrained_ec: "./model/pretrained/deeplab-18"
   pretrained_epoch: 0
   PIXEL_MEANS:
   - 103.06

diff --git a/init.sh b/init.sh
@@ -2,7 +2,7 @@
 
 mkdir -p ./output
 mkdir -p ./external/mxnet
-mkdir -p ./model/pretrained_model
+mkdir -p ./model/pretrained
 
 cd lib/bbox
 python setup_linux.py build_ext --inplace