Eye On Intelligence Challenge - a Birds-Eye view - Not Quite Final Blog

19 Nov 2024

I was in Rhode Island setting up equipment for a field experiment last week for work so I'm a little late with this project. I'll summarize my progress, however, the project is not quite complete.

After several hours of trying to install various versions of TensorFlow on the original Pynq-Z1 3.0.1 image with Python 3.10 without success, I reverted to Pynq-Z1 2.7.0 with Python 3.8 and finally TensorFlow installed with all its dependencies. Trying to start the Jupyter Lab Notebook failed and I'm still trying to figure that problem out.

So.....my hardware is partially installed in the box, the software kinda works on the PC but I haven't been able to run it on the Arty-Z7 yet because I haven't figured out how to get the Jupyter Notebook to run the iPython kernal.

I will continue to work to finish this project even though the deadline has passed and post the finished project when I get it done.

Congratulations to the contestants that finished.

Top Comments

vmate 7 months ago +1

Do you know which object recognition model you will use? Or is it something custom? If you need help with training, I have the hardware to do it quickly, and also the software environment already set up…

DAB 7 months ago

I hope you do finish the project.

I was looking forward to seeing how well it does with real birds.
- Cancel
- Vote Up 0 Vote Down
- Sign in to reply
- More
- Cancel
rsc 7 months ago in reply to embeddedguy

Right now I think I have an issue with the ipython kernel. I'm going to try and re-install it. Jupyter lab will start, but the PC doesn't recognize the Arty board for some reason.
- Cancel
- Vote Up 0 Vote Down
- Sign in to reply
- More
- Cancel
rsc 7 months ago in reply to vmate
The model is working on the PC, but I'm still trying to get it to work on the Arty-Z7

model = tf.keras.Sequential([
tf.keras.layers.Rescaling(1./255),
tf.keras.layers.Conv2D(32, 3, activation='relu'),
tf.keras.layers.MaxPooling2D(),
tf.keras.layers.Conv2D(32, 3, activation='relu'),
tf.keras.layers.MaxPooling2D(),
tf.keras.layers.Conv2D(32, 3, activation='relu'),
tf.keras.layers.MaxPooling2D(),
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dense(num_classes)
])
```
Epoch 10/10
4/4 ━━━━━━━━━━━━━━━━━━━━ 3s 589ms/step - accuracy: 0.8428 - loss: 0.5521 - val_accuracy: 0.6875 - val_loss: 0.7710
```
- Cancel
- Vote Up 0 Vote Down
- Sign in to reply
- More
- Cancel
vmate 7 months ago

Do you know which object recognition model you will use? Or is it something custom?

If you need help with training, I have the hardware to do it quickly, and also the software environment already set up, message me and I'll happily help.
- Cancel
- Vote Up +1 Vote Down
- Sign in to reply
- More
- Cancel
embeddedguy 7 months ago

It's okay. If you want to try CV2 and YOLO it works well with PYNQ. Tensorflow and/or Pytorch may create dependency related problems..
- Cancel
- Vote Up 0 Vote Down
- Sign in to reply
- More
- Cancel