Hey guys!! If you'd like to learn more about OpenCV in Python, I'm doing new FIRE series where we do that by MAKING MEMES, check it out: ua-cam.com/video/Fe32So9a1ts/v-deo.html
Hey guy🙌, From the coco dataset, I can annotate and separate only images containing objects "Man, traffic light, road sign, stop sign, bench, backpack, umbrella, handbag, suitcase"?????
Thank you so much, Connor! I know right? I was low-key thinking looking for goats & teddy bears wouldn't work at all :DD I guess, it really takes that sort of ridiculous combination popping up to demonstrate the vastness of a dataset, right? Should be added as a bench mark to all segmentation / OD datasets, imo :DD
I have question sir I am doing the same thing as you do what i do with train image of coco dataset it size is 428 *600 but bounding box and label are 10. How i deal with that cardinality ambiguit?,
@@IvanGoncharovAI Dear Ivan i normalize it, but it gives error your x size is 427 and y is 5 i passed 427*500 size image to model with 2 bounding boxes any 2 label it gives error x size is different with y ? I am fully stuck with it how model train with images having different size class label for multi object detection in one image. :(
Hello, I want to convert a COCO dataset annotation to a binary mask. The problem I have is that after conversion with pycocotools method annToMask nearly all the labels have the same colour and the segmentation network recognises it as one label. Is there a fix to that so that every label has a different colour. Please help
Hello! Haven't had experience with converting COCO stuff to masks, so can't say for sure. If you give me more details, maybe I could help, but nothing useful that I can say now
I understand the part of getting the bounding boxes of train images, but what if we want the instances segmentation coordinates (x and y) of the test images?
how do i need to change the code exaclty if i want to have multiple classes in the desired pictures? lets say humans and carrots. do i just add new for loops to the big iteration loop?
In the context of the COCO dataset, it's possible to get images with classes that are present in the dataset. I am not sure if ID cards or helmets are there but you can check on their website
Hello Iván, all good? I hope so. Ivan maybe I have my last question on 5his topic, but maybe is a tricky one and I was looking on internet and found that many people do it but doesn't explain why. I was testing a network that I trained on 480x480 resolution, but didn't get quite good results. So I decide to reduce the resolution quite a bit, little by little actually (every 32 pixels I reduce) until I found a good result to me on a resolution of 320x320. So I want to understand why by reducing the resolution on which yolo works, it gets a better result on recall and precision.
There's no way to say for sure what happens but it may have something to do with the improved efficiency of training, and maybe it could also reduce overfitting since there are less features to overfit on with lesser resolution
@@IvanGoncharovAI mmm thank you Ivan. But I forget to said this. I trained the network at resolution of 480*480, but at the validation process, in the cfg file I changed the resolution to 320*320, getting better results. So I was thinking that maybe the network as it is trained at a higher resolution, can maybe get better results during validation using a lesser resolution, due to the fact it is easier to it to recognize more simple details than the ones it saw during training. It is just my guest, I would like to know yours. Sorry to be so annoying, but I saw again the comment I left you and I didn't specify my self well at the first instance
Hey guys!! If you'd like to learn more about OpenCV in Python, I'm doing new FIRE series where we do that by MAKING MEMES, check it out: ua-cam.com/video/Fe32So9a1ts/v-deo.html
Yo Ivan!! Thanks man, love the content. I love how non repetitive your content is and how you break down the concepts
Thank you!
Really helpful video, thankyou Ivan!
Just the video I am looking for. Kudos to your explanation. Subscribed
Glad it was useful, Gowtam!
Hey guy🙌, From the coco dataset, I can annotate and separate only images containing objects "Man, traffic light, road sign, stop sign, bench, backpack, umbrella, handbag, suitcase"?????
I'm new to coco, what I want to say is, is it possible to use only certain objects?
Really cool video! Didn't know you could search for images that combine categories like zebra and cake, lol! Image at 3:24...
Thank you so much, Connor! I know right? I was low-key thinking looking for goats & teddy bears wouldn't work at all :DD I guess, it really takes that sort of ridiculous combination popping up to demonstrate the vastness of a dataset, right? Should be added as a bench mark to all segmentation / OD datasets, imo :DD
Thanks for kindly explaining.
Thanks for watching!
I have question sir I am doing the same thing as you do what i do with train image of coco dataset it size is 428 *600 but bounding box and label are 10. How i deal with that cardinality ambiguit?,
I mean, I don't remember exactly which are which, but I think the classes and bboxes are normalized to be like percentages values
@@IvanGoncharovAI Dear Ivan i normalize it, but it gives error your x size is 427 and y is 5 i passed 427*500 size image to model with 2 bounding boxes any 2 label it gives error x size is different with y ? I am fully stuck with it how model train with images having different size class label for multi object detection in one image. :(
how you upload the dataset I mean how you keep your google collab from reconnecting while uploading
I'd say first upload the dataset to Google Drive and then you can copy it from there to Colab, check out the 2nd Colab video in the playlist
Thanks! Great video!
Thanks!
Hi! Thanks for sharing this, I just want to split the caption file instead of instances, so I really appreciate it if you can help me with this.
Hello,
I want to convert a COCO dataset annotation to a binary mask. The problem
I have is that after conversion with pycocotools method annToMask nearly all the labels have the same colour and the segmentation network recognises it as one label. Is there a fix to that so that every label has a different colour. Please help
Hello! Haven't had experience with converting COCO stuff to masks, so can't say for sure. If you give me more details, maybe I could help, but nothing useful that I can say now
@@IvanGoncharovAI Thanks, I already found a libary with a method which can convert the annotations to binary masks it's called pycocotools
hlo brother can you help me with a project,i wanted to add bounding box in a project called knww osteothritis detection project
hey man! what do you need help with?
Sir, how can I evaluate it? Just would like to know the mean average precision of YOLO using COCO dataset.
Usually yolo frameworks have a command for evaluation on COCO dataset
I understand the part of getting the bounding boxes of train images, but what if we want the instances segmentation coordinates (x and y) of the test images?
Unfortunately this video doesn't cover that, Houssam. But you can look up some stuff on the COCO website
how do i need to change the code exaclty if i want to have multiple classes in the desired pictures? lets say humans and carrots. do i just add new for loops to the big iteration loop?
Yeah, you can just update the indices in the loops to include the objects that you want, shouldn't be too hard looking at the code
hey there! unzipping of train2017 is taking too much time. what should i do?
Prolly waiting
Is it possible to include more objects like helmets,ID cards etc.?
In the context of the COCO dataset, it's possible to get images with classes that are present in the dataset. I am not sure if ID cards or helmets are there but you can check on their website
thanks... very useful...
Thanks for watching!
Hello Iván, all good? I hope so.
Ivan maybe I have my last question on 5his topic, but maybe is a tricky one and I was looking on internet and found that many people do it but doesn't explain why. I was testing a network that I trained on 480x480 resolution, but didn't get quite good results. So I decide to reduce the resolution quite a bit, little by little actually (every 32 pixels I reduce) until I found a good result to me on a resolution of 320x320. So I want to understand why by reducing the resolution on which yolo works, it gets a better result on recall and precision.
There's no way to say for sure what happens but it may have something to do with the improved efficiency of training, and maybe it could also reduce overfitting since there are less features to overfit on with lesser resolution
@@IvanGoncharovAI mmm thank you Ivan. But I forget to said this. I trained the network at resolution of 480*480, but at the validation process, in the cfg file I changed the resolution to 320*320, getting better results. So I was thinking that maybe the network as it is trained at a higher resolution, can maybe get better results during validation using a lesser resolution, due to the fact it is easier to it to recognize more simple details than the ones it saw during training. It is just my guest, I would like to know yours. Sorry to be so annoying, but I saw again the comment I left you and I didn't specify my self well at the first instance