new model

the export in perceptilabs now also includes preprocessing, so normalisation is removed from code now
birdstream · Sep 9, 2021 · 38189e2 · 38189e2
1 parent bb890e9
commit 38189e2
Show file tree

Hide file tree

Showing 5 changed files with 157 additions and 142 deletions.
diff --git a/inference.py b/inference.py
@@ -60,7 +60,7 @@ async def predict(bild: bytes = File(...)):
     for b in range(block_rows):
         for a in range(block_cols):
             pred_in = img_array[b,a,:,:].reshape(1, 224, 224, 3) # pick out the block to be predicted
-            predict = np.asarray(model(pred_in / norm)['target']).reshape(224, 224, 3) # make the prediction and reshape output (omit batch)
+            predict = np.asarray(model(pred_in)['target']).reshape(224, 224, 3) # make the prediction and reshape output (omit batch)
             img_array_out[(b * 208):(b * 208) + 208, (a * 208):(a * 208) + 208, :] = predict[8:216, 8:216, :] # put the predicted block in it's right place in the new array
     img_out = tf.keras.preprocessing.image.array_to_img(img_array_out) # convert array to image
     img_out = img_out.crop((0, 0, W, H)) # crop out the padding

diff --git a/model/keras_metadata.pb b/model/keras_metadata.pb
diff --git a/model/saved_model.pb b/model/saved_model.pb
diff --git a/model/variables/variables.data-00000-of-00001 b/model/variables/variables.data-00000-of-00001
diff --git a/model/variables/variables.index b/model/variables/variables.index