Added requirements.txt file for version compatibility

.gitignore

@@ -3,8 +3,12 @@
 *.swo
 *.swm
 
+*.pth
+
 *.lprof
 
 .cache
 __pycache__
 src/__pycache__
+
+

README.md

@@ -6,6 +6,8 @@ from the paper [*Quo Vadis, Action Recognition? A New Model and the Kinetics Dat
 
 The original (and official!) tensorflow code can be found [here](https://github.com/deepmind/kinetics-i3d/).
 
+To install the required libraries: `pip install -r requirements.txt`
+
 The heart of the transfer is the `i3d_tf_to_pt.py` script
 
 Launch it with `python i3d_tf_to_pt.py --rgb` to generate the rgb checkpoint weight pretrained from ImageNet inflated initialization.
@@ -14,6 +16,8 @@ To generate the flow weights, use `python i3d_tf_to_pt.py --flow`.
 
 You can also generate both in one run by using both flags simultaneously `python i3d_tf_to_pt.py --rgb --flow`.
 
+If you have errors regarding Protobuf versions, try `PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python python i3d_tf_to_pt.py --rgb --flow`.
+
 Note that the master version requires PyTorch 0.3 as it relies on the recent addition of ConstantPad3d that has been included in this latest release.
 
 If you want to use pytorch 0.2 checkout the branch pytorch-02 which contains a simplified model with even padding on all sides (and the corresponding pytorch weight checkpoints).

i3d_pt_demo.py

@@ -9,10 +9,13 @@
 
 
 def run_demo(args):
+
+    device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
+
     kinetics_classes = [x.strip() for x in open(args.classes_path)]
 
     def get_scores(sample, model):
-        sample_var = torch.autograd.Variable(torch.from_numpy(sample).cuda())
+        sample_var = torch.autograd.Variable(torch.from_numpy(sample).to(device))
         out_var, out_logit = model(sample_var)
         out_tensor = out_var.data.cpu()
 
@@ -30,7 +33,7 @@ def get_scores(sample, model):
         i3d_rgb = I3D(num_classes=400, modality='rgb')
         i3d_rgb.eval()
         i3d_rgb.load_state_dict(torch.load(args.rgb_weights_path))
-        i3d_rgb.cuda()
+        i3d_rgb.to(device)
 
         rgb_sample = np.load(args.rgb_sample_path).transpose(0, 4, 1, 2, 3)
         out_rgb_logit = get_scores(rgb_sample, i3d_rgb)
@@ -40,7 +43,7 @@ def get_scores(sample, model):
         i3d_flow = I3D(num_classes=400, modality='flow')
         i3d_flow.eval()
         i3d_flow.load_state_dict(torch.load(args.flow_weights_path))
-        i3d_flow.cuda()
+        i3d_flow.to(device)
 
         flow_sample = np.load(args.flow_sample_path).transpose(0, 4, 1, 2, 3)
         out_flow_logit = get_scores(flow_sample, i3d_flow)