1. Question 1 — How to scale pixel values to 0–1?

• ❌ normalize parameter

• ❌ Normalization layer

• ✅ Rescaling layer (tf.keras.layers.Rescaling())

• ❌ TensorFlow does it automatically

• ❌ rescale parameter

Explanation:
Use:

---

2. Question 2 — How does image_dataset_from_directory assign labels?

• ✅ Based on the directory each image is contained in

• ❌ TensorFlow reads contents

• ❌ Based on filename

• ❌ Manual labeling required

Explanation:
Each folder name becomes a class label.

---

3. Question 3 — Effects of reducing image resolution

Correct answers (multiple):

• ✅ Training is faster

• ❌ You no longer need to rescale

• ❌ Training results may differ (TRUE but you missed it)

• ✅ You lose information

✔️ Correct choices: Training is faster + Training results may differ + Info loss

---

4. Question 4 — Meaning of input_shape = (300, 300, 3)

• ❌ 3 bytes

• ❌ 300 images

• ❌ 3 conv layers

• ✅ Each image is 300×300 with 3 color channels (RGB)

---

5. Question 5 — Training accuracy 1.000 but low validation

• ✅ You’re overfitting on your training data

• ❌ Overfitting validation

• ❌ Underfitting validation

• ❌ No risk

Explanation:
High train accuracy + low val accuracy = classic overfitting.

---

6. Question 6 — How to set target image resolution?

• ❌ target_size

• ❌ Rescaling

• ✅ image_size parameter

• ❌ training_size

Explanation:

---

🧾 Summary Table

---