Fixed issue #1 (output channel bug)

.gitignore

@@ -31,4 +31,4 @@ data/
 *.ckpt*
 
 #   Ignore test images
-test.*
+*.jpg

predict.py

@@ -157,12 +157,12 @@ def predict(img_list, model_path=None, thresh=0.05, is_name=False, cpu_only=True
         model.restore_sess(model_path)
 
     #   get the last stage's result
-    pred_map = model.sess.run(model.output, feed_dict={model.img: _input / 255.0})[:, -1]
+    pred_map = model.sess.run(model.output, feed_dict={model.img: _input / 255.0})[:, -1, :, :, :-1]
     if debug:
         np.save('pred.npy', pred_map)
 
-    j = -1 * np.ones((len(_img_list), pred_map.shape[-1]-1, 2))
-    w = np.zeros((len(_img_list), pred_map.shape[-1]-1))
+    j = -1 * np.ones((len(_img_list), pred_map.shape[-1], 2))
+    w = np.zeros((len(_img_list), pred_map.shape[-1]))
     for idx in range(len(_img_list)):
         #   re-project heatmap to origin size
 
@@ -192,10 +192,7 @@ def predict(img_list, model_path=None, thresh=0.05, is_name=False, cpu_only=True
 if __name__=='__main__':
     """ Demo of Using the model API
     """
-    img_names = ['000061164.jpg', '000078951.jpg', '000094304.jpg', '000099899.jpg',
-                '000065339.jpg', '000085370.jpg', '000094342.jpg', '000109154.jpg',
-                '000071686.jpg', '000090584.jpg', '000099186.jpg','000111209.jpg'
-                ]
+    img_names = ['test.jpg']
     #   input must be greater than one
     assert len(img_names) >= 1
     j = predict(img_names, 'model/model.ckpt-99', debug=True, is_name=True)

valPCK.py

@@ -96,14 +96,15 @@ def compute_distance(model, dataset, metric='PCKh', debug=False):
         j_gt = np.array(j_list)
 
         #   estimate by network
-        j_dt,_ = predict.predict(im_list, model=model, thresh=0.0)
+        j_dt,_ = predict.predict(im_list, model=model, thresh=0.0, debug=True)
         w = np.transpose(np.hstack((np.expand_dims(w,1), np.expand_dims(w,1))), axes=[0,2,1])
+        assert j_dt.shape == j_gt.shape
 
-        for n in range(len(Global.joint_list)):
+        for n in range(j_dt.shape[0]):
             err[_iter*paral+n] = np.linalg.norm(w[n]*(j_gt[n,:,:]-j_dt[n,:,-1::-1]),axis=1) / np.linalg.norm(j_gt[n,normJ_a,:]-j_gt[n,normJ_b,:], axis=0)
         print "[*]\tTemp Error is ", np.average(err[_iter*paral:_iter*paral+paral], axis=0)
         if debug:
-            break
+            return err
 
     aver_err = np.average(err)
     print "[*]\tAverage PCKh Normalised distance is ", aver_err
@@ -123,5 +124,5 @@ def compute_distance(model, dataset, metric='PCKh', debug=False):
     model.BuildModel()
     model.restore_sess('model/model.ckpt-99')
 
-    dist = compute_distance(model, dataset, metric='PCKh', debug=True)
+    dist = compute_distance(model, dataset, metric='PCKh')
     visualize_accuracy(dist, start=0.01, end=0.5, showlist=[0,1,4,5,10,11,14,15])