1f3da600c20e3376b0bb396bba482b1e90b7883c,i3nception_tf.py,,,#,34

Before Change

        out3d = unitpy(input_3d_var)
        filter_idx = 5
        py_out = py_to_tf_reshape(
            out3d.data.numpy())[0][0][:, :, filter_idx].copy()
        tf_out = tf_out3dsample[0][0][:, :, filter_idx].copy()
        import pdb
        pdb.set_trace()

After Change

    // init = tf.global_variables_initializer()
    // sess.run(init)
    for i, (input_2d, target) in enumerate(loader):
        input_2d = torch.from_numpy(input_2d.numpy())
        target = target.cuda()
        target_var = torch.autograd.Variable(target)

        // Pytorch forward pass
        input_3d = input_2d.clone().unsqueeze(2).repeat(1, 1, frame_nb, 1, 1)
        input_3d_var = torch.autograd.Variable(input_3d)

        feed_dict = {}
        input_3d_tf = input_3d.numpy().transpose(0, 2, 3, 4, 1)  //
        feed_dict[rgb_input] = input_3d_tf

        // Get output
        tf_out3dsample = sess.run(rgb_logits, feed_dict=feed_dict)
        out_tf_np = tf_out3dsample.transpose((0, 4, 1, 2, 3))
        out_tf = torch.from_numpy(out_tf_np)

        unit_name_tf = "RGB/inception_i3d/Conv3d_1a_7x7/"

        // Create state_dict
        state_dict = {}
        i3nception.load_conv3d(state_dict, "", sess, unit_name_tf)
        unitpy.eval()
        unitpy.load_state_dict(state_dict)
        out_pt = unitpy(input_3d_var).data
        out_pt_np = out_pt.numpy()
        filter_idx = 0

        // Plot slices
        filter_idx = 0
        img_tf = out_tf_np[0][filter_idx][0]
        img_pt = out_pt_np[0][filter_idx][0]

        max_v = max(img_tf.max(), img_pt.max())
        min_v = min(img_tf.min(), img_pt.min())
        plt.subplot(2, 2, 1)

Instances