Optical flow concept in Computer Vision - Model Metrics & Evaluation

Metrics & Evaluation - Optical flow concept

Which metric matters for Optical Flow and WHY

Optical flow measures how pixels move between video frames. The key metric is End-Point Error (EPE). It shows the average distance between the predicted pixel movement and the true movement. A smaller EPE means the model tracks motion more accurately. This matters because good motion tracking helps in video analysis, robotics, and self-driving cars.

Confusion matrix or equivalent visualization

Optical flow is a regression task, not classification, so no confusion matrix applies. Instead, we use a visual comparison:

    Frame 1 pixels:  (x, y)
    Frame 2 pixels:  (x + u, y + v)  where (u, v) is the flow vector

    True flow vector: (u_true, v_true)
    Predicted flow vector: (u_pred, v_pred)

    End-Point Error (EPE) = sqrt((u_true - u_pred)**2 + (v_true - v_pred)**2)
    

Lower EPE means better prediction of pixel movement.

Precision vs Recall tradeoff (or equivalent) with examples

In optical flow, the tradeoff is between accuracy and smoothness of the flow field.

• High accuracy but low smoothness: The model predicts pixel movement very close to true values but may produce noisy or jagged flow, which can confuse downstream tasks.
• High smoothness but low accuracy: The flow looks smooth and clean but may miss small or fast movements, reducing usefulness.

Example: For a self-driving car, accurate detection of fast-moving objects is critical, so accuracy is prioritized. For video stabilization, smooth flow is more important.

What "good" vs "bad" metric values look like for Optical Flow

Good: EPE close to 0 pixels (e.g., < 1 pixel) means the predicted motion is very close to the true motion.

Bad: EPE greater than 5 pixels means the predicted motion is often far from the true motion, causing errors in motion understanding.

Visual checks also help: good flow fields show smooth, consistent arrows matching object movement; bad flow fields look noisy or random.

Common pitfalls in Optical Flow metrics

• Ignoring occlusions: Pixels that disappear or appear between frames can cause wrong error calculations.
• Using only average EPE: It can hide large errors in small regions; median or percentile errors give more insight.
• Overfitting to training data: Model may predict well on known scenes but fail on new videos.
• Data leakage: Using future frames or ground truth flow in training can give unrealistically low errors.

Self-check question

Your optical flow model has an average EPE of 0.5 pixels on test videos but shows noisy flow arrows in fast-moving areas. Is it good?

Answer: The low EPE suggests good average accuracy, but noisy flow in fast areas means the model struggles with quick motion. It may need improvement in those regions for reliable use.

Key Result

End-Point Error (EPE) is the key metric; lower EPE means better optical flow accuracy.

Practice

1. What does optical flow represent in computer vision?

easy

A. The pattern of apparent motion of objects between two images

B. The color distribution in a single image

C. The brightness level of an image

D. The shape of objects detected in an image

Solution

1. Step 1: Understand the meaning of optical flow

   Optical flow describes how pixels move from one image frame to the next, showing motion.

2. Step 2: Compare options to definition

   Only The pattern of apparent motion of objects between two images correctly describes motion between images, while others describe unrelated image properties.

3. Final Answer:

   The pattern of apparent motion of objects between two images -> Option A

4. Quick Check:

   Optical flow = motion pattern [OK]

Hint: Optical flow = pixel motion between frames [OK]

Common Mistakes:

• Confusing optical flow with color or brightness
• Thinking optical flow detects object shapes
• Assuming optical flow works on single images

2. Which OpenCV function is commonly used to compute dense optical flow using the Farneback method?

easy

A. cv2.calcOpticalFlowPyrLK()

B. cv2.findContours()

C. cv2.threshold()

D. cv2.calcOpticalFlowFarneback()

Solution

1. Step 1: Recall OpenCV functions for optical flow

   Farneback method is implemented in OpenCV as cv2.calcOpticalFlowFarneback().

2. Step 2: Eliminate unrelated functions

   Other options relate to tracking points, contours, or thresholding, not Farneback optical flow.

3. Final Answer:

   cv2.calcOpticalFlowFarneback() -> Option D

4. Quick Check:

   Farneback method = calcOpticalFlowFarneback [OK]

Hint: Farneback method uses cv2.calcOpticalFlowFarneback() [OK]

Common Mistakes:

• Confusing Farneback with Lucas-Kanade method
• Choosing functions unrelated to optical flow
• Mixing up function names with similar prefixes

3. Given the following Python code using OpenCV, what is the shape of the output variable flow?

import cv2
prev_img = cv2.imread('frame1.png', 0)
next_img = cv2.imread('frame2.png', 0)
flow = cv2.calcOpticalFlowFarneback(prev_img, next_img, None, 0.5, 3, 15, 3, 5, 1.2, 0)

medium

A. A 3D array with shape (height, width, 2)

B. A 2D array with shape (height, width)

C. A 1D array with length equal to number of pixels

D. A scalar value representing average flow

Solution

1. Step 1: Understand output of Farneback optical flow

   The function returns a 3D array where each pixel has a 2D flow vector (x and y components).

2. Step 2: Match output shape to options

   This means the output shape is (height, width, 2), representing flow vectors per pixel.

3. Final Answer:

   A 3D array with shape (height, width, 2) -> Option A

4. Quick Check:

   Flow shape = (H, W, 2) [OK]

Hint: Optical flow output has two channels for x and y motion [OK]

Common Mistakes:

• Assuming output is 2D grayscale image
• Thinking output is a single scalar
• Confusing flow vector dimensions

4. You wrote this code to compute optical flow but get an error:

flow = cv2.calcOpticalFlowFarneback(prev_img, next_img, 0, 0.5, 3, 15, 3, 5, 1.2, 0)

What is the likely cause of the error?

medium

A. The window size parameter is missing

B. The images must be color, not grayscale

C. The third argument must be None, not 0

D. The pyramid scale parameter is too high

Solution

1. Step 1: Check the third argument of calcOpticalFlowFarneback

   The third argument is the output flow array or None to create it internally; passing 0 causes an error.

2. Step 2: Verify correct usage

   Passing None lets OpenCV allocate the output; 0 is invalid and causes a type error.

3. Final Answer:

   The third argument must be None, not 0 -> Option C

4. Quick Check:

   Third argument = None [OK]

Hint: Use None as third argument to allocate flow output [OK]

Common Mistakes:

• Passing 0 instead of None for output flow
• Using color images instead of grayscale
• Misunderstanding parameter order

5. You want to track a fast-moving object in a video using optical flow. Which approach helps improve accuracy?

hard

A. Use a larger window size and fewer pyramid levels

B. Use a smaller window size and increase pyramid levels

C. Convert images to color before computing flow

D. Apply thresholding before computing flow

Solution

1. Step 1: Understand parameters affecting optical flow accuracy

   Smaller window size captures finer motion details; more pyramid levels help track fast motion at multiple scales.

2. Step 2: Evaluate options for fast-moving object tracking

   Use a smaller window size and increase pyramid levels correctly suggests smaller window and more pyramid levels to improve tracking accuracy.

3. Final Answer:

   Use a smaller window size and increase pyramid levels -> Option B

4. Quick Check:

   Smaller window + more pyramids = better fast motion tracking [OK]

Hint: Smaller window + more pyramids track fast motion better [OK]

Common Mistakes:

• Using large windows that blur fast motion
• Ignoring pyramid levels for scale changes
• Preprocessing images incorrectly