this link<\/a>.<\/p><\/blockquote>\nStep 3: Loading the model and studying its input and output<\/h1>\n
Now that we have the model and our development environment ready, the next step is to create a Python snippet that allows us to load this model and run inferencing with it.<\/p>\n
Here\u2019s what such a snippet might look like:<\/p>\n
import numpy as np\nimport tensorflow as tf\n\n# Load TFLite model and allocate tensors.\ninterpreter = tf.contrib.lite.Interpreter(model_path=\"exposure.tflite\")\n\n# Get input and output tensors.\ninput_details = interpreter.get_input_details()\noutput_details = interpreter.get_output_details()\n\ninterpreter.allocate_tensors()\n\n# input details\nprint(input_details)\n# output details\nprint(output_details)<\/pre>\nHere, we first load the downloaded model and then get the input and output tensors from it.<\/p>\n
Next, we print the input and outputs tensors we obtained earlier.
\nIf you run the code, this is what the output might look like:<\/p>\n
Input: [{'name': 'image', 'index': 0, 'shape': array([ 1, 224, 224, 3], dtype=int32), 'dtype': <class 'numpy.uint8'>, 'quantization': (0.007874015718698502, 128)}]<\/strong><\/span>Output: [{'name': 'scores', 'index': 173, 'shape': array([1, 3], dtype=int32), 'dtype': <class 'numpy.uint8'>, 'quantization': (0.00390625, 0)}]<\/strong><\/span><\/pre>\nHere, the model takes the input at index 0, and the type of input is an array of shape [1,224,224,3], which basically means that it takes a single image of size 224 x 224 in RGB format.<\/p>\n
The model gives the output at index 173, and the shape of the output array is [1,3], which essentially means we\u2019ll be getting the scores for each of our labels. For the purposes of my project, I\u2019ve trained the model to classify an input image as underexposed, overexposed, or good\u2014so the output has the shape [1,3].<\/p>\n
If I trained my model to detect n<\/strong> labels, then the output shape would have been [1,n].<\/p>\nNow that we have the input and output type and shapes of the model, let\u2019s load an image and run it through the TensorFlow Lite model.<\/p>\n
Step 4: Reading an image and passing it to the TFLite model<\/h1>\n
Up next, we\u2019ll use Pathlib to iterate through a folder containing some images that we\u2019ll be running inference on. We\u2019ll then read each image with OpenCV, resize it to 224×224, and pass it to our model.<\/p>\n
Once done, we\u2019ll print the file name and the output for that file:<\/p>\n
import numpy as np\nimport tensorflow as tf\nimport cv2\nimport pathlib\n\n# Load TFLite model and allocate tensors.\ninterpreter = tf.contrib.lite.Interpreter(model_path=\"exposure.tflite\")\n\n# Get input and output tensors.\ninput_details = interpreter.get_input_details()\noutput_details = interpreter.get_output_details()\n\ninterpreter.allocate_tensors()\n\n# input details\nprint(input_details)\n# output details\nprint(output_details)\n\nfor file in pathlib.Path(folder_path).iterdir():\n\n # read and resize the image\n img = cv2.imread(r\"{}\".format(file.resolve()))\n new_img = cv2.resize(img, (224, 224))\n\n # input_details[0]['index'] = the index which accepts the input\n interpreter.set_tensor(input_details[0]['index'], [new_img])\n\n # run the inference\n interpreter.invoke()\n\n # output_details[0]['index'] = the index which provides the input\n output_data = interpreter.get_tensor(output_details[0]['index'])\n\n print(\"For file {}, the output is {}\".format(file.stem, output_data))<\/pre>\nUpon running this code, here\u2019s what the output might look like:<\/p>\n
For file DSC00117, the output is [[ 75 164 17]]\nFor file DSC00261, the output is [[252 2 2]]\nFor file DSC00248, the output is [[252 2 2]]\nFor file DSC00116, the output is [[ 21 210 25]]\nFor file DSC00128, the output is [[ 9 112 136]]\nFor file DSC00114, the output is [[203 42 12]]<\/span><\/pre>\nAs you can see, the output contains 3 values, which add up to 256. These represent the scores for the image, which indicate whether it\u2019s underexposed, good, or overexposed, respectively.<\/p>\n
To get your model\u2019s predicted output, simply fetch the index with the greatest score and map it with your label.<\/p>\n
For example, in the first image in the output above, the highest score is the second element in the array, which maps with the label \u201cgood<\/strong>\u201d; whereas, the second image is \u201cunderexposed<\/strong>\u201d and the second to last is \u201coverexposed<\/strong>\u201d.<\/p>\nTo further improve model performance, you can batch the requests as well. Let\u2019s see how to do this.<\/p>\n
Step 5: Batching requests for better performance<\/h1>\n
Batching with TensorFlow essentially means running a bunch of inference requests at once instead of running them one by one.<\/p>\n
This will result in a reduced latency in our model.<\/p>\n
Before we go ahead and batch the requests, we need to decide upon the batch size. This might vary from machine to machine, and setting a very large batch size might result in an out of memory exception.<\/p>\n
To make things simple, we will set the batch size as 50. Here\u2019s what the above example with batching looks like:<\/p>\n
import numpy as np\nimport tensorflow as tf\nimport cv2\nimport pathlib\n\n# Load TFLite model and allocate tensors.\ninterpreter = tf.contrib.lite.Interpreter(model_path=\"exposure.tflite\")\n\n# Get input and output tensors.\ninput_details = interpreter.get_input_details()\noutput_details = interpreter.get_output_details()\n\ninterpreter.allocate_tensors()\n\n# input details\nprint(input_details)\n# output details\nprint(output_details)\n\nimages = []\n\nfor index, file in enumberate(pathlib.Path(folder_path).iterdir()):\n\n # read and resize the image\n img = cv2.imread(r\"{}\".format(file.resolve()))\n new_img = cv2.resize(img, (224, 224))\n\n images.append(new_img)\n\n if index == 49:\n interpreter.set_tensor(input_details[0]['index'], images)\n # run the inference\n interpreter.invoke()\n # output_details[0]['index'] = the index which provides the input\n output_data = interpreter.get_tensor(output_details[0]['index'])\n # clear the list\n images.clear()<\/pre>\nBut upon running the code above, you might see an error, which looks as follows:<\/p>\n
Traceback (most recent call last):\n File \"test.py\", line 48, in <module>\n interpreter.set_tensor(input_details[0]['index'], input_data)\n File \"\/Users\/harshitdwivedi\/Desktop\/tf_env\/lib\/python3.7\/site-packages\/tensorflow\/lite\/python\/interpreter.py\", line 175, in set_tensor\n self._interpreter.SetTensor(tensor_index, value)\n File \"\/Users\/harshitdwivedi\/Desktop\/tf_env\/lib\/python3.7\/site-packages\/tensorflow\/lite\/python\/interpreter_wrapper\/tensorflow_wrap_interpreter_wrapper.py\", line 136, in SetTensor\n return _tensorflow_wrap_interpreter_wrapper.InterpreterWrapper_SetTensor(self, i, value)\nValueError: Cannot set tensor: Dimension mismatch<\/strong><\/span><\/pre>\nThe reason for this is a mismatch in the shape of the input tensor vs the shape of the input that we\u2019re providing.<\/p>\n
The input that the model accepts is [1, 224, 224, 3], whereas the input we\u2019re providing is [50, 224, 224, 3]. To fix this, we can simply resize our input before running inference on it.<\/p>\n
Here\u2019s what the code for doing so looks like:<\/p>\n
import numpy as np\nimport tensorflow as tf\nimport cv2\nimport pathlib\n\n# Load TFLite model and allocate tensors.\ninterpreter = tf.contrib.lite.Interpreter(model_path=\"exposure.tflite\")\n\n# Get input and output tensors.\ninput_details = interpreter.get_input_details()\noutput_details = interpreter.get_output_details()\n\n# input details\nprint(input_details)\n# output details\nprint(output_details)\n\nimages = []\n\nfor index, file in enumberate(pathlib.Path(folder_path).iterdir()):\n\n # read and resize the image\n img = cv2.imread(r\"{}\".format(file.resolve()))\n new_img = cv2.resize(img, (224, 224))\n\n images.append(new_img)\n\n if index == 49:\n # resize the input tensor\n interpreter.resize_tensor_input(input_details[0]['index'],[len(images), 224, 224, 3])\n interpreter.allocate_tensors()\n interpreter.set_tensor(input_details[0]['index'], images)\n # run the inference\n interpreter.invoke()\n # output_details[0]['index'] = the index which provides the input\n output_data = interpreter.get_tensor(output_details[0]['index'])\n # clear the list\n images.clear()<\/pre>\nBefore running inference, we can also resize the input so that it accepts as many images as there are in the current batch of images.<\/p>\n
On running the code above, this is what you might get as the output:<\/p>\n
[[ 75 164 17], [252 2 2], [252 2 2], [ 21 210 25], ..., [203 42 12], [ 9 112 136]]<\/span><\/pre>\nThe length of this array would be the same as your batch size (here 50); so if you maintain another list of file names, you can reference the score for each image easily!<\/p>\n<\/div>\n<\/div>\n<\/div>\n\n\n\n
<\/div>\n\n\n\n
\n
\n
\n
And that\u2019s it! While not always the model effective solution, TFLite models are nonetheless an extremely viable alternative when it comes to running your models on edge hardware, or if the model\u2019s latency is a core concern for your app!<\/p>\n
In the next part of this series, I\u2019ll be covering how we can do the same for object detection TF Lite models, in order to locate and track detected objects in an image. Stay tuned for that!<\/p>\n
Thanks for reading! If you enjoyed this story, please <\/em>click the <\/em><\/strong>\ud83d\udc4f button<\/em><\/strong> <\/em>and share it <\/em><\/strong>to help others find it! Feel free to leave a comment <\/em>\ud83d\udcac below.<\/em><\/p>\nHave feedback? Let\u2019s connect <\/em>on Twitter<\/em><\/strong><\/a>.<\/em><\/p>\n<\/div>\n<\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"Photo by Guillaume de Germain on Unsplash Following up on my earlier blogs on running edge models in Python, this fifth blog in the series of Training and running Tensorflow models will explore how to run a TensorFlow Lite image classification model in Python. If you haven\u2019t read my earlier post on model training for this task, you […]<\/p>\n","protected":false},"author":56,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"customer_name":"","customer_description":"","customer_industry":"","customer_technologies":"","customer_logo":"","footnotes":""},"categories":[7],"tags":[],"coauthors":[159],"class_list":["post-6873","post","type-post","status-publish","format-standard","hentry","category-tutorials"],"yoast_head":"\n
Running TensorFlow Lite Image Classification Models in Python - Comet<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Running TensorFlow Lite Image Classification Models in Python\" \/>\n<meta property=\"og:description\" content=\"Photo by Guillaume de Germain on Unsplash Following up on my earlier blogs on running edge models in Python, this fifth blog in the series of Training and running Tensorflow models will explore how to run a TensorFlow Lite image classification model in Python. If you haven\u2019t read my earlier post on model training for this task, you […]\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/\" \/>\n<meta property=\"og:site_name\" content=\"Comet\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/cometdotml\" \/>\n<meta property=\"article:published_time\" content=\"2023-07-19T18:01:40+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2025-04-24T17:15:10+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/miro.medium.com\/v2\/resize:fit:2500\/1*aEOXmVBaGdZvuoK6uFi7mQ.jpeg\" \/>\n<meta name=\"author\" content=\"Harshit Dwivedi\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@Cometml\" \/>\n<meta name=\"twitter:site\" content=\"@Cometml\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Harshit Dwivedi\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"8 minutes\" \/>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"Running TensorFlow Lite Image Classification Models in Python - Comet","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/","og_locale":"en_US","og_type":"article","og_title":"Running TensorFlow Lite Image Classification Models in Python","og_description":"Photo by Guillaume de Germain on Unsplash Following up on my earlier blogs on running edge models in Python, this fifth blog in the series of Training and running Tensorflow models will explore how to run a TensorFlow Lite image classification model in Python. If you haven\u2019t read my earlier post on model training for this task, you […]","og_url":"https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/","og_site_name":"Comet","article_publisher":"https:\/\/www.facebook.com\/cometdotml","article_published_time":"2023-07-19T18:01:40+00:00","article_modified_time":"2025-04-24T17:15:10+00:00","og_image":[{"url":"https:\/\/miro.medium.com\/v2\/resize:fit:2500\/1*aEOXmVBaGdZvuoK6uFi7mQ.jpeg","type":"","width":"","height":""}],"author":"Harshit Dwivedi","twitter_card":"summary_large_image","twitter_creator":"@Cometml","twitter_site":"@Cometml","twitter_misc":{"Written by":"Harshit Dwivedi","Est. reading time":"8 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/#article","isPartOf":{"@id":"https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/"},"author":{"name":"Harshit Dwivedi","@id":"https:\/\/www.comet.com\/site\/#\/schema\/person\/3bd50ca837ac698f7041652196a3456d"},"headline":"Running TensorFlow Lite Image Classification Models in Python","datePublished":"2023-07-19T18:01:40+00:00","dateModified":"2025-04-24T17:15:10+00:00","mainEntityOfPage":{"@id":"https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/"},"wordCount":1181,"publisher":{"@id":"https:\/\/www.comet.com\/site\/#organization"},"image":{"@id":"https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/#primaryimage"},"thumbnailUrl":"https:\/\/miro.medium.com\/v2\/resize:fit:2500\/1*aEOXmVBaGdZvuoK6uFi7mQ.jpeg","articleSection":["Tutorials"],"inLanguage":"en-US"},{"@type":"WebPage","@id":"https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/","url":"https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/","name":"Running TensorFlow Lite Image Classification Models in Python - Comet","isPartOf":{"@id":"https:\/\/www.comet.com\/site\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/#primaryimage"},"image":{"@id":"https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/#primaryimage"},"thumbnailUrl":"https:\/\/miro.medium.com\/v2\/resize:fit:2500\/1*aEOXmVBaGdZvuoK6uFi7mQ.jpeg","datePublished":"2023-07-19T18:01:40+00:00","dateModified":"2025-04-24T17:15:10+00:00","breadcrumb":{"@id":"https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/#primaryimage","url":"https:\/\/miro.medium.com\/v2\/resize:fit:2500\/1*aEOXmVBaGdZvuoK6uFi7mQ.jpeg","contentUrl":"https:\/\/miro.medium.com\/v2\/resize:fit:2500\/1*aEOXmVBaGdZvuoK6uFi7mQ.jpeg"},{"@type":"BreadcrumbList","@id":"https:\/\/www.comet.com\/site\/blog\/running-tensorflow-lite-image-classification-models-in-python\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.comet.com\/site\/"},{"@type":"ListItem","position":2,"name":"Running TensorFlow Lite Image Classification Models in Python"}]},{"@type":"WebSite","@id":"https:\/\/www.comet.com\/site\/#website","url":"https:\/\/www.comet.com\/site\/","name":"Comet","description":"Build Better Models Faster","publisher":{"@id":"https:\/\/www.comet.com\/site\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.comet.com\/site\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/www.comet.com\/site\/#organization","name":"Comet ML, Inc.","alternateName":"Comet","url":"https:\/\/www.comet.com\/site\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.comet.com\/site\/#\/schema\/logo\/image\/","url":"https:\/\/www.comet.com\/site\/wp-content\/uploads\/2025\/01\/logo_comet_square.png","contentUrl":"https:\/\/www.comet.com\/site\/wp-content\/uploads\/2025\/01\/logo_comet_square.png","width":310,"height":310,"caption":"Comet ML, Inc."},"image":{"@id":"https:\/\/www.comet.com\/site\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/cometdotml","https:\/\/x.com\/Cometml","https:\/\/www.youtube.com\/channel\/UCmN63HKvfXSCS-UwVwmK8Hw"]},{"@type":"Person","@id":"https:\/\/www.comet.com\/site\/#\/schema\/person\/3bd50ca837ac698f7041652196a3456d","name":"Harshit Dwivedi","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.comet.com\/site\/#\/schema\/person\/image\/54c8c9192b5afd9f7101fbd03edda390","url":"https:\/\/www.comet.com\/site\/wp-content\/uploads\/2023\/08\/o34DyZS__400x400-96x96.jpg","contentUrl":"https:\/\/www.comet.com\/site\/wp-content\/uploads\/2023\/08\/o34DyZS__400x400-96x96.jpg","caption":"Harshit Dwivedi"},"url":"https:\/\/www.comet.com\/site\/blog\/author\/harshit-dwivedigmail-com\/"}]}},"_links":{"self":[{"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/posts\/6873","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/users\/56"}],"replies":[{"embeddable":true,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/comments?post=6873"}],"version-history":[{"count":1,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/posts\/6873\/revisions"}],"predecessor-version":[{"id":15601,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/posts\/6873\/revisions\/15601"}],"wp:attachment":[{"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/media?parent=6873"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/categories?post=6873"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/tags?post=6873"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/www.comet.com\/site\/wp-json\/wp\/v2\/coauthors?post=6873"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
![\"\"](\"https:\/\/miro.medium.com\/v2\/resize:fit:640\/format:webp\/1*FVZjRpJ6Wxdefl2Z_CMKfg.png)
<\/picture>![\"\"](\"https:\/\/miro.medium.com\/v2\/resize:fit:1000\/1*FVZjRpJ6Wxdefl2Z_CMKfg.png\")
<\/figure>