Day 6: Matplotlib: plotting distributions and images
Seeing your data before you trust it
Matplotlib is the plumbing every other Python plotting library sits on top of. You need exactly two things from it for this roadmap: plotting distributions (is this data sane?) and displaying images (did this CV step actually work, or did it silently produce garbage?). Both habits pay off immediately in Stage 0 and constantly afterward — a training curve in Stage 1, a segmentation mask in Stage 2.
import matplotlib.pyplot as plt
fig, axes = plt.subplots(1, 2, figsize=(10, 4))
axes[0].hist(orders["price"], bins=30)
axes[0].set_title("Price distribution")
axes[1].bar(return_rate_by_category.index, return_rate_by_category.values)
axes[1].set_title("Return rate by category")
plt.tight_layout()
plt.savefig("sanity_check.png")import cv2
img = cv2.imread("garment.jpg")
img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) # OpenCV loads BGR — see Day 8
plt.imshow(img_rgb)
plt.axis("off")
plt.title("Loaded garment photo")
plt.savefig("check.png")Why this is worth a whole day
The single most common classical-CV bug is a pipeline that runs without errors but silently produces the wrong output — a background-removal mask that's inverted, a resize that squashed the aspect ratio. imshow at every pipeline stage is how you catch this in seconds instead of during Day 14's 20-photo validation.
Key terms
- Histogram
- A plot of how often values fall into each of a set of bins — the standard way to sanity-check a numeric distribution.
- imshow
- Matplotlib's function for rendering a 2D or 3D array as an image — the core debugging tool for any CV pipeline.
A background-removal function runs without any errors, but the output looks wrong when you actually view it. What should you have done earlier?
