Examples: vision and tflite simpletests

modules/tflite/examples/tflite_simpletest.py

@@ -0,0 +1,56 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+"""Run a TensorFlow Lite model: predict a sine wave.
+
+Loads a tiny model that learned sin(x) for x from 0 to 2π, asks it for a
+few values, and prints each prediction beside math.sin. The model works in
+8-bit integers, so each x is scaled into the model's units on the way in
+and back out again on the way out.
+
+Needs: a MicroPython firmware built with tflite (it's opt-in: name it in
+the build), and the model on the board. It is esp-vision's
+models/tflite/sine/sine.tflite, from https://github.com/espressif/esp-vision.
+
+Run it:
+    mpremote cp sine.tflite :
+    mpremote run tflite_simpletest.py
+
+Tested on: a Waveshare ESP32-P4-WIFI6-Touch-LCD-4B on MicroPython 1.30
+preview.
+"""
+
+import math
+
+import tflite
+
+# Change these to match your setup.
+MODEL = "sine.tflite"
+POINTS = 8
+
+# Load the model, and read how it scales numbers into 8-bit integers.
+model = tflite.Model(MODEL)
+in_scale, in_zero = model.input_scale[0], model.input_zero_point[0]
+out_scale, out_zero = model.output_scale[0], model.output_zero_point[0]
+print("Loaded %s: %d bytes, %d bytes of working memory" % (MODEL, model.len, model.ram))
+
+
+def predict(x):
+    """The model's guess at sin(x)."""
+
+    # The model asks for its input by handing over a buffer to fill.
+    def fill(buf, shape, dtype):
+        buf[0] = max(-128, min(127, round(x / in_scale) + in_zero)) & 0xFF
+
+    # The output is one row of one number, in the model's units.
+    output = model.predict([fill])[0]
+    return (output[0][0] - out_zero) * out_scale
+
+
+# Ask for a few points across one cycle, and compare with math.sin.
+for i in range(POINTS):
+    x = 2 * math.pi * i / (POINTS - 1)
+    print("sin(%.2f) is %+.3f; the model says %+.3f" % (x, math.sin(x), predict(x)))
+
+# Free the model's memory.
+model.deinit()

modules/vision/examples/vision_simpletest.py

@@ -0,0 +1,47 @@
+# SPDX-FileCopyrightText: 2026 Brad Barnett
+#
+# SPDX-License-Identifier: MIT
+"""Find a coloured shape in a picture.
+
+Draws a red rectangle and a blue circle on a blank picture with the image
+module, then asks find_blobs for everything red and prints where it is.
+It saves the picture as a JPEG, so you can look at it yourself. With a
+camera, the same find_blobs call finds red things in its pictures.
+
+Needs: a MicroPython firmware built with vision (it's opt-in: name it in
+the build), on an ESP32-S3 or ESP32-P4. No camera or display.
+
+Run it:
+    mpremote run vision_simpletest.py
+    mpremote cp :shapes.jpg .
+
+Tested on: a Waveshare ESP32-P4-WIFI6-Touch-LCD-4B on MicroPython 1.30
+preview.
+"""
+
+import image
+
+# Change these to match your setup.
+WIDTH = 320
+HEIGHT = 240
+FILENAME = "shapes.jpg"
+
+# Red, as find_blobs sees colour: lightness from 30 to 100, and well towards
+# red on both of the LAB colour space's colour axes (15 to 127).
+RED = (30, 100, 15, 127, 15, 127)
+
+# Draw a red rectangle and a blue circle on a black picture.
+picture = image.Image(WIDTH, HEIGHT, image.RGB565)
+picture.clear()
+picture.draw_rectangle(100, 60, 80, 50, color=(255, 0, 0), fill=True)
+picture.draw_circle(40, 40, 20, color=(0, 0, 255), fill=True)
+
+# Find every red patch of at least 100 pixels.
+for blob in picture.find_blobs([RED], pixels_threshold=100, area_threshold=100):
+    print("Found red at x=%d, y=%d, %d wide and %d high" % blob.rect())
+
+# Save the picture as a JPEG.
+jpeg = picture.to_jpeg(quality=85, copy=True)
+with open(FILENAME, "wb") as f:
+    f.write(jpeg.bytearray())
+print("Saved", FILENAME)