Update app.py

app.py

@@ -12,42 +12,6 @@ def classify_img(img):
   pred,idx,probs=learn.predict(img)
   return dict(zip(categories, map(float, probs)))
 
-def gradcam(img_create):
-    pred,idx,probs=learn.predict(img)
-    
-    x,= first(dls_augmented.test_dl([img]))
-    hook_output = Hook()
-    hook = learn.model[0].register_forward_hook(hook_output.hook_func)
-    with torch.no_grad(): output = learn.model.eval()(x)
-    act = hook_output.stored[0]
-    hook.remove()
-
-    input_size=act.shape[0]
-    out_size=learn.model[1][-1].in_features
-    kernel_size=act.shape[1]
-    a=act
-    new_act=tensor(np.zeros((out_size,kernel_size,kernel_size)))
-    sum=tensor(np.zeros((1,kernel_size,kernel_size)))
-    for i in range(0,input_size,4):
-        sum=tensor(np.zeros((1,kernel_size,kernel_size))).to('cuda')
-    for j in range(i,i+4):
-        sum=sum+act[j,:,:]
-    new_act[int(i/4),:,:]=sum/4
-    cam_map = torch.einsum('ck,kij->cij', learn.model[1][-1].weight, new_act.to('cuda'))
-
-    pred_scores=[ round(elem, 2)*100 for elem in learn.predict(img_create)[2].tolist() ]
-    preds = dict(sorted({labels[i]: [ round(elem, 2)*100 for elem in learn.predict(img_create)[2].tolist() ][i] for i in range(len(labels))}.items(), key=lambda x:x[1], reverse=True))
-
-    x_dec = TensorImage(dls_augmented.train.decode((x,))[0][0])
-    fig,ax = plt.subplots()
-    x_dec.show(ctx=ax)
-    ax.imshow(cam_map[1].detach().cpu(), alpha=0.6, extent=(0,128,128,0), interpolation='bilinear', cmap='magma');
-    im="gcam.jpg"
-    fig.savefig(im)
-    
-    return (dict(zip(categories, map(float, probs))), im)
-    
-
 image=gr.inputs.Image(shape=(128,128))
 label=gr.outputs.Label()
 examples_=[]
@@ -59,6 +23,6 @@ examples=["filibe-1-1.jpg",
           "varna-1-1.jpg"]
 
 
-demo = gr.Interface(fn=gradcam, inputs=image, outputs=[label,image], examples=examples)
+demo = gr.Interface(fn=classify_img, inputs=image, outputs=[label,image], examples=examples)
 
 demo.launch(inline=False)