Wildfire classification with MobileNetV2-based model #1315
Replies: 2 comments 3 replies
-
|
Hi, chunki! |
Beta Was this translation helpful? Give feedback.
-
|
Hi Mengzou01!
I am glad that the repo may help you. My internet access will be very limited until next Monday, but here you are some comments:
- Characterize: it never worked for me either. It is supossed to provide more optimized FIFO sizes, but I tried several times and stopped it after more than 24 hours running without results. I did not change the default number of cores variable, which I beleive it is 4, so it may affect.
I recommend the following configuration, using "full_build_json_folding":
```
my_auto_fifo_depths = True
my_auto_fifo_strategy = "largefifo_rtlsim"
my_folding_config_file = './folding_config/manual_folding_config.json'
```
- Calculating FIFO sizes takes a long time, maybe 2 or 3 hours, and the full build took almost 10 hours in a powerful machine, although sharing other worloads. So it depends on your PC, but be pacient.
- On the other hand, "final_hw_config.json" is the file produced by the [step_set_fifo_depths](https://github.com/Xilinx/finn/blob/main/src/finn/builder/build_dataflow_steps.py#L533), before large FIFOs are splitted. I guess that some attributes, as "numInputVectors", are inserted during the [step_apply_folding_config](https://github.com/Xilinx/finn/blob/main/src/finn/builder/build_dataflow_steps.py#L445), which are not compatible with QONNX allowed attributes. Maybe you could overcome this issue with some coding. FINN researchers could also confirm if "final_hw_config.json" can be used directly as it is to avoid calculating FIFO sizes, so configuring ```my_auto_fifo_depths=False```.
Best regards!
…On Wed, Apr 16, 2025 at 2:44 PM Mengzou01 ***@***.***> wrote:
Hi, chunki!
I'm very excited to have come across your MobileNetV2-based wildfire
classification project. However, I'm encountering an error while rebuilding.
I successfully ran the "estimates" phase using
finn_build_config_pynq_z1.ipynq. But, when executing either the
"full_build" or "full_build_json_folding" process, an error occurs.
I changed parameter:"my_auto_fifo_strategy" from "largefifo_rtlsim" into
"characterize"(stuck too long time if largefifo_rtlsim), and changed
parameter:"my_folding_config_file" to the config_file your supported in
foler "folding_config" which named "final_hw_config.json".
Any suggestions to resolve this issue would be greatly appreciated. Thanks!
If the following files will be helpful:
build_dataflow.log
<https://github.com/user-attachments/files/19777185/build_dataflow.log>
logfile.log
<https://github.com/user-attachments/files/19777165/logfile.log>
image.png (view on web)
<https://github.com/user-attachments/assets/ee28d4ee-7d6a-4151-b820-0b8b6b161d68>
—
Reply to this email directly, view it on GitHub
<#1315 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/BQPYMJXQH6IIEGHIGS7CZET2ZZGCTAVCNFSM6AAAAAB2YT7M7GVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTEOBVGQ2DKMI>
.
You are receiving this because you authored the thread.Message ID:
***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
Hi chunki, |
Beta Was this translation helpful? Give feedback.
-
|
Hi, step_set_fifo_depths, step_create_stitched_ip and step_synthesize_bitfile take a long time, several hours in my machine each, so be patient. If you just want to check the driver, you can find it in the PYNQ folder, inside Best regards! PD: sorry for the format, I do not know why the markdown is not working. |
Beta Was this translation helpful? Give feedback.
-
|
Hi, |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Hi everyone!
As part of my research on AI at the edge, developed at ETSIT, Universidad Politécnica de Madrid, I'd like to share this project: MobileNetV2-based model for wildfire classification.
It is divided into three main parts:
Many thanks to the researchers at AMD/Xilinx for their work and for making Brevitas and FINN open-source!
Beta Was this translation helpful? Give feedback.
All reactions