As you can see the image, something that has not been clear for me is that if you enter more than one word in the image search, the model can categorize by words, not the entire search.
After train the model, I think the error_rate is somewhat high, but it works quite well for being my first training of a model.
Hello everyone, After first lesson, I trained a model to recognize if a photo is of cheetah or leopard.
Here is the kaggle notebook link: leopard_or_cheetah | Kaggle
I got pretty good loss values, but the model still make some mistakes
Hi Folks!
I have completed Lecture 2 Assignment. The project I chose was classifying the paintings made by Picasso and Monet (I couldn’t think of any crazy project ). Deployed it in HuggingFace Spaces and also wrote a blog (setup my blog using Quarto & Github Pages) about the project with source code. Happy feelings!
I’m excited to share my latest project with you all: a pet and predator classifier! Check it out Here
Together we grow!! Looking forward to your feedback
It’s great joy for me being part of this community.
Tonight (over here), I’m glad to share that with the understanding gotten from the course taught by @jeremy and greatly referencing the “Is it a bird?” notebook, I’ve been able to build a smooth “Is there a roadblock ahead?” computer vision model that is greatly able to determine if there’s a roadblock ahead in no time, both at night and day.
The model which is fine-tuned on a dataset of “roadblock and free road photos” uses the pre-trained “resnet18” model.
Resnet18 did amazingly well and Kaggle is so kind with the free GPUs offering of 30hrs. Lol, and I’m just getting started. Cheers.
Please, my friends, I’d appreciate your comments and feedback on this.
Thanks, here’s the link; Is there a roadblock ahead?
Hey everyone! I’m new here but excited to be part of the community. I’ve got a unique challenge and I’m hoping to get some advice on which Fastai model might be the best fit. I’m working with structured spreadsheet data to explore the relationship between food intake (with details like macros) and bowel movements, trying to identify trigger foods. The twist is the delayed reaction - up to 48 hours between eating and the effects. So, it’s not straightforward.
Here’s the lowdown:
Data: Two tables. One with food and nutrient intake, and another describing bowel movements (e.g., thickness, fragmentation).
Objective: Predict ‘good’ vs ‘bad’ bowel movements based on food intake, accounting for the up to 48-hour delay. Ideally this is a Regression model with a scale of the degree of fragmentation, thickness, etc. But to start a simple binary “good” or “bad” would suffice.
Data Size: 6 months worth.
Given the delay in reaction and the structured nature of my data, I’m scratching my head on how to model this. Any suggestions on models or a particular approach within fastai that could handle this delayed effect scenario effectively?
I studied aeronautics in university and found it confusing. Having just completed Part 1 of this course, I see a different path. Complex things can be simplified. We develop a feel for what’s happening behind the Greek letters. Things start to click.
Hello, I am a newbie here. I tried doing classification problem to detect if there is a spoofing signal.
Spoofing signal can be thought as fake signal. When an external transmitter transmits similar signal features to existing true signals, spoofing occurs if there is no safeguard in the receiver side against it. As a result, it could drag a receiver to a wrong information (e.g. position, time). As input data, I used the correlation values of the input signal and replica signal generated from my receiver. It looks like this in 3D:
In the two above images, you can see that if spoofing occurs, we can see two peaks; while if not, only one peak is present. However, these are not really my input datasets. I tried to focus more on spoofing where in you can only see one peak such as these:
In the two images above, we can only see one peak but the second image has actually spoofed signal. The obvious difference can be found in the values of those peaks: with spoofing, lower correlation value; without spoofing, higher correlation value.
I converted this to images and set them as input to my model (which I just followed from the introduction) like this:
True signal:
in the path variable, it points to my data folder that contains two subfolders: training and valid. Does the dataloaders use both subfolders?
is the model training based on the intensity values in the images? For example, in the images i gave above, true signal only has higher correlation value, while with spoofed signal correlation value is lower.
Thanks for your feedback! Please let me know if there are unclear points. This is also my first time posting in the forums! Thanks for all your work.
I think I can answer your first question: yes, by default, the recurse parameter of get_image_files is set to True so it will grab all image file paths in all subfolders. If you wanted to make your training/validation split based on those subfolders you could use GrandParentSplitter instead of RandomSplitter.
Regarding your second question: the model is likely learning different patterns about true and spoofed signals, intensity being one of then, but also other features like the shape of the signal, its position, and so on. In those particular images you showed the true signal is brighter, but unless that is always the case, the model will have to learn other features than the color of the signal pixels. Perhaps someone else can give a more detailed answer specific to your use case.
I asked that second question because I am worried that position is being taken into account. Of course in different time shift, the position of the peak may be different. The model won’t be performing best at that instance. That’s also the reason why I think there is a better way to create my input datasets. Pixel sizes are also big about 1800 x 1800 i think. But still need to figure out
I appreciate your time and effort to share your insights. Thanks.
Hope you all are doing great. I am new to ML and I come from a Software Engineering background. I am currently pursuing a Master’s in Software Engineering and have always been excited about learning ML. I came across an article explaining why one should learn this course as a beginner. So far, loving it. Jeremy’s style of teaching is what I liked the most.
I completed the first chapter and implemented an Animal Classifier, though not super clear with every line of code. But as per Jeremy’s suggestion, I am moving forward with the next topics. Hopefully, I learn and understand the topics well in this course by the end.
Hello, I finished lesson 1 (animal classifier) and modified it to a new, harder, classification problem: is it a duck or a goose?
I managed to get it to about 80% accuracy with 300 images of each type and increasing the epochs of training - not bad. More data, and cleaner data would probably make this even more accurate but it was a very interesting exercise.
Thanks to the course designers, looking forward to doing the rest of the course.
Hey, I am newbie here i finished lesson 1 and i tried the kaggle if its a bird but it showed many errors so i checked resent post by other members and changed the code a bit and done my work here it is :
I hope ill learn more from class and as well as from members
I built a simple football player (using a FIFA world cup 2022 dataset from Kaggle) classifier and used the predict method to compare which professional player closely resembles your own. You can upload an image via file or webcam. The image returned uses search_images_ddg method to source the predicted players image so it is not always an accurate image.
The classifier was trained using resnet50 in the vision_learner.
Please feel free to try this and peruse the code.
I am still quite new to this field so I would appreciate any feedback as you seem fit (both the bad and good!). Thank you!
I am on the 2 Chapter of the book , i tried but i am unable to use this … "To download images with Bing Image Search, sign up at [Microsoft Azure] for a free account. You will be given a key, which you can copy and enter in a cell as follows (replacing ‘XXX’ with your key and executing it):
). Deployed it in HuggingFace Spaces and also wrote a blog (setup my blog using Quarto & Github Pages) about the project with source code. Happy feelings!
Check it out
I’m new here but excited to be part of the community. I’ve got a unique challenge and I’m hoping to get some advice on which Fastai model might be the best fit. I’m working with structured spreadsheet data to explore the relationship between food intake (with details like macros) and bowel movements, trying to identify trigger foods. The twist is the delayed reaction - up to 48 hours between eating and the effects. So, it’s not straightforward.
i finished lesson 1 and i tried the kaggle if its a bird but it showed many errors so i checked resent post by other members and changed the code a bit and done my work here it is :