diff --git a/models/ResNet50/pokedex_ResNet50.onnx b/models/ResNet50/pokedex_ResNet50.onnx
new file mode 100644
index 0000000000000000000000000000000000000000..0730f25196dc2551aefbbec09ce54e69200fe9bb
Binary files /dev/null and b/models/ResNet50/pokedex_ResNet50.onnx differ
diff --git a/python/convert_onnx.py b/python/convert_onnx.py
index 846a0b8b66d6be8285407ce952242c4d0400c942..a90d6d1b941da7cd367a300c02670a3393ddd9ff 100644
--- a/python/convert_onnx.py
+++ b/python/convert_onnx.py
@@ -1,6 +1,7 @@
import tensorflow as tf
import tf2onnx
import argparse
+import keras
# --- WHAT ? ---
parser = argparse.ArgumentParser(description="WHAT ?!")
@@ -20,16 +21,16 @@ elif args.model == "2":
size_2=(1, 256, 256, 3)
# --- Load Sequential model ---
-seq_model = tf.keras.models.load_model(h5_path, compile=False)
+seq_model = keras.models.load_model(h5_path, compile=False)
# --- Create input layer with same shape ---
-inputs = tf.keras.Input(shape=size, name="input")
+inputs = keras.Input(shape=size, name="input")
# --- Call the Sequential model as a function ---
outputs = seq_model(inputs)
# --- Wrap in Functional model ---
-model = tf.keras.Model(inputs=inputs, outputs=outputs)
+model = keras.Model(inputs=inputs, outputs=outputs)
# --- Convert to ONNX ---
spec = (tf.TensorSpec(size_2, tf.float32, name="input"),)
diff --git a/python/hailo_compile.py b/python/hailo_compile.py
deleted file mode 100644
index 2f25e2c51e3d0b8edd916bf30457d1d53b2551c7..0000000000000000000000000000000000000000
--- a/python/hailo_compile.py
+++ /dev/null
@@ -1,59 +0,0 @@
-import os
-import subprocess
-import argparse
-import onnx
-
-parser = argparse.ArgumentParser(description="Convert, parse, optimize and compile model for Hailo.")
-parser.add_argument("--model", choices=["1", "2"], required=True, help="1 = ResNet50, 2 = Xception")
-parser.add_argument("--calib", default="--use-random-calib-set", help="Calibration data path or '--use-random-calib-set'")
-args = parser.parse_args()
-
-def get_onnx_io_names(onnx_path):
- model = onnx.load(onnx_path)
- input_name = model.graph.input[0].name
- output_name = model.graph.output[0].name
- return input_name, output_name
-
-# Paths
-if args.model == "1":
- base_model_path = "../models/ResNet50"
- model = "pokedex_ResNet50"
-elif args.model == "2":
- base_model_path = "../models/Xception"
- model = "pokedex_Xception"
-
-onnx_path = os.path.join(base_model_path, f"pokedex_{model}.onnx")
-har_path = os.path.join(base_model_path, f"{model}.har")
-optimized_har_path = os.path.join(base_model_path, f"{model}_optimized.har")
-hef_path = os.path.join(base_model_path, f"{model}.hef")
-
-# Node names
-start_node, end_node = get_onnx_io_names(onnx_path)
-print(f"-- Using start_node: {start_node}, end_node: {end_node}")
-
-# Step 1: Parse
-print(f"-- Parsing {onnx_path}...")
-subprocess.run([
- "hailo", "parser", "onnx", onnx_path,
- "--start-node-names", start_node,
- "--end-nodes-names", end_node,
- "--hw-arch", "hailo8l"
-])
-
-# Step 2: Optimize
-print(f"-- Optimizing to {optimized_har_path}...")
-optimize_cmd = [
- "hailo", "optimize", har_path,
- "--hw-arch", "hailo8l",
- "--use-random-calib-set"
-]
-subprocess.run(optimize_cmd)
-
-# Step 3: Compile
-print(f"-- Compiling to {hef_path}...")
-subprocess.run([
- "hailo", "compiler", optimized_har_path,
- "--hw-arch", "hailo8l",
- "--performance",
-])
-print("-- Done.")
diff --git a/python/pokedex_test.py b/python/pokedex_test.py
index 011ab2951b78163e7f483a2f91350766749eab0a..6a3b9668832f8856e6b3fd5ce29b68eff5a50c55 100644
--- a/python/pokedex_test.py
+++ b/python/pokedex_test.py
@@ -1,5 +1,5 @@
-from tensorflow import keras
import tensorflow as tf
+import keras
import matplotlib.pyplot as plt
import numpy as np
import os
@@ -7,74 +7,70 @@ import random
import json
import argparse
-# --- WHAT ? ---
-parser = argparse.ArgumentParser(description="WHAT ?!")
+# --- Parse CLI arguments ---
+parser = argparse.ArgumentParser(description="Test trained Pokémon model.")
parser.add_argument("--model", choices=["1", "2"], required=True, help="1 = ResNet50, 2 = Xception")
args = parser.parse_args()
-# Paths
+# --- Paths ---
if args.model == "1":
h5_path = "../models/ResNet50/pokedex_ResNet50.h5"
json_path = "../models/ResNet50/class_names.json"
- size = (224,224)
+ size = (224, 224)
elif args.model == "2":
h5_path = "../models/Xception/pokedex_Xception.h5"
json_path = "../models/Xception/class_names.json"
- size = (256,256)
+ size = (256, 256)
+
+base_path = "../Combined_Dataset"
-# --- Load class names from JSON ---
+# --- Load class names ---
with open(json_path, "r") as f:
class_names = json.load(f)
- class_names = [class_names[i] for i in range(len(class_names))] # convert to list
+ class_names = [class_names[i] for i in range(len(class_names))]
-# --- Load trained model ---
-model = keras.models.load_model(h5_path)
+# --- Load model (NO COMPILE to avoid 'reduction=auto' bug) ---
+model = keras.models.load_model(h5_path, compile=False)
-# --- Paths ---
-base_path = "../Combined_Dataset"
-
-# --- Prepare 2x2 Plot ---
+# --- 2x2 Random Image Test ---
plt.figure(figsize=(10, 10))
for i in range(4):
- # Pick random class and image
- random_class = random.choice(class_names)
- class_folder = os.path.join(base_path, random_class)
- random_image = random.choice([
+ # Pick random Pokémon class & image
+ true_class = random.choice(class_names)
+ class_folder = os.path.join(base_path, true_class)
+ img_file = random.choice([
f for f in os.listdir(class_folder)
if f.lower().endswith(('.png', '.jpg', '.jpeg'))
])
- img_path = os.path.join(class_folder, random_image)
+ img_path = os.path.join(class_folder, img_file)
- # --- Load & Preprocess Image ---
- img = keras.utils.load_img(img_path, target_size=size) # resize to match model input
+ # Load and preprocess image
+ img = keras.utils.load_img(img_path, target_size=size)
img_array = keras.utils.img_to_array(img)
- img_array = img_array / 255.0 # normalize if your model expects it
- img_array = tf.expand_dims(img_array, 0)
+ img_array = tf.expand_dims(img_array, 0) # [1, height, width, 3]
- # --- Predict ---
+ # Predict
predictions = model.predict(img_array, verbose=0)
- probabilities = tf.nn.softmax(predictions[0])
- predicted_class_index = np.argmax(probabilities)
- predicted_label = class_names[predicted_class_index]
- confidence = 100 * probabilities[predicted_class_index]
-
- top_5_indices = np.argsort(probabilities)[-5:][::-1]
- print("\nTop 5 predictions:")
- for idx in top_5_indices:
- print(f"{class_names[idx]:<20}: {probabilities[idx]:.4f}")
-
+ probabilities = tf.nn.softmax(predictions[0]).numpy()
+ predicted_index = np.argmax(probabilities)
+ predicted_label = class_names[predicted_index]
+ confidence = 100 * probabilities[predicted_index]
+ is_correct = predicted_label == true_class
- # Compare with actual
- is_correct = predicted_label == random_class
+ # Show top 5
+ print(f"\n Image: {img_file} | True: {true_class}")
+ print("-- Top 5 predictions:")
+ for idx in np.argsort(probabilities)[-5:][::-1]:
+ print(f"{class_names[idx]:<20}: {probabilities[idx]*100:.2f}%")
- # --- Plot ---
+ # Plot
ax = plt.subplot(2, 2, i + 1)
plt.imshow(img)
plt.axis("off")
plt.title(
f"Pred: {predicted_label}\n"
- f"True: {random_class}\n"
+ f"True: {true_class}\n"
f"{'YES' if is_correct else 'NO'} | {confidence:.1f}%",
fontsize=10
)