Training routine tested

README.md

@@ -50,11 +50,11 @@ flow = computeFlow(im)
 ```
 
 ## Training
-Training sequentially is faster than training end-to-end since you need to learn small number of parameters at each level. To train a level `N`, we need the trained models at levels `1` to `N-1`.
+Training sequentially is faster than training end-to-end since you need to learn small number of parameters at each level. To train a level `N`, we need the trained models at levels `1` to `N-1`. You also initialize the model with a pretrained model at `N-1`.
 
-E.g. To train level 3, we need trained models at `L1` and `L2`. 
+E.g. To train level 3, we need trained models at `L1` and `L2`, and we initialize it  `modelL2_3.t7`.
 ```bash
-th main.lua -fineWidth 128 -fineHeight 96 -level 3 -netType volcon -cache checkpoint -data FLYING_CHAIRS_DIR -L1 models/modelL1_3 -L2 modelsL2_3
+th main.lua -fineWidth 128 -fineHeight 96 -level 3 -netType volcon -cache checkpoint -data FLYING_CHAIRS_DIR -L1 models/modelL1_3.t7 -L2 models/modelL2_3.t7 -retrain models/modelL2_3.t7
 ```
 ## Timing Benchmarks
 Our timing benchmark is set up on Flying chair dataset. To test it, you need to download

checkpoint/TueJan3116:50:362017/train.log

@@ -0,0 +1,2 @@
+avg loss (train set)	% top1 accuracy (train set)	
+ 2.1416e+00	 0.0000e+00	

donkey.lua

@@ -62,10 +62,10 @@ local modelL1path, modelL2path, modelL3path, modelL4path
 local down1, down2, down3, down4, up2, up3, up4
 local warpmodel2, warpmodel3, warpmodel4
 
-modelL1path = paths.concat('models', opt.L1)
-modelL2path = paths.concat('models', opt.L2)
-modelL3path = paths.concat('models', opt.L3)
-modelL4path = paths.concat('models', opt.L4)
+modelL1path = opt.L1
+modelL2path = opt.L2
+modelL3path = opt.L3
+modelL4path = opt.L4
 
 if opt.level > 1 then
    -- Load modelL1

opts.lua

@@ -39,10 +39,10 @@ function M.parse(arg)
     cmd:option('-weightDecay',     5e-4, 'weight decay')
     cmd:option('-optimizer',       'adam', 'adam or sgd')
     ---------- Model options ----------------------------------
-    cmd:option('-L1', 'modelL1_4.t7', 'Trained Level 1 model')
-    cmd:option('-L2', 'modelL2_4.t7', 'Trained Level 2 model')
-    cmd:option('-L3', 'modelL3_4.t7', 'Trained Level 3 model')
-    cmd:option('-L4', 'modelL4_4.t7', 'Trained Level 4 model')
+    cmd:option('-L1', 'models/modelL1_4.t7', 'Trained Level 1 model')
+    cmd:option('-L2', 'models/modelL2_4.t7', 'Trained Level 2 model')
+    cmd:option('-L3', 'models/modelL3_4.t7', 'Trained Level 3 model')
+    cmd:option('-L4', 'models/modelL4_4.t7', 'Trained Level 4 model')
 
     cmd:option('-netType',     'volcon', 'Lua network file')
     cmd:option('-retrain',     'none', 'provide path to model to retrain with')