00:35:11 Jonathan Shao: As Tavis pointed out you might have to restart kernel as well 00:35:16 Jonathan Shao: Then run the code block 00:35:17 Sean Kearney (USDA-ARS): Jonathan's method worked for me. I had to add --user at the end though 00:37:41 Yanbo Huang: Still not working 00:49:24 Jonathan Shao: do more steps help in accuracy 00:56:33 Tavis Anderson: It’s brutally slow on my MacBook - sorry. I think I’ll be mostly following along with your examples! 00:57:33 Aaron Szczepanek: saxophone 00:57:45 Aaron Szczepanek: 776 00:58:55 Sean Kearney (USDA-ARS): fly (308) 01:06:01 Aaron Szczepanek: the deepest convolution 01:06:25 Aaron Szczepanek: second deepest lol 01:07:16 Aaron Szczepanek: random 01:07:44 Aaron Szczepanek: coool 01:09:50 Jonathan Shao: My computer has also decided not to run any of the examples will be following along with onscreen examples as well 01:21:46 Sean Kearney (USDA-ARS): too slow on my computer. don't have GPU 01:21:50 Sean Kearney (USDA-ARS): How about fly again? 01:21:57 Sean Kearney (USDA-ARS): (308) 01:22:16 Sean Kearney (USDA-ARS): oh wait - no it doesn't! 01:23:27 Sean Kearney (USDA-ARS): Gradcam did run - took maybe 30 secs. not too bad. 01:24:05 Aaron Szczepanek: what is gradcam using to make these? 01:25:54 Aaron Szczepanek: so looking at all layers 01:26:57 Aaron Szczepanek: cool tho! 01:53:21 Alex Styer (he/him): Which module includes struct (for struct.unpack)? I’m getting an error that struct is not defined 01:54:56 Tavis Anderson: Import struct worked for me 01:56:02 Alex Styer (he/him): Yeah import struct was missing from my first block (I downloaded via curl for the ceres instructions—maybe an older version of the notebook?) 01:56:07 Alex Styer (he/him): Works now! thanks! 01:57:16 Scott Tsukuda: Thanks Alex! 02:14:08 WGMeikle: Could you specify exactly where the 'import struct' line should go? 02:14:12 Aaron Szczepanek: we could use what we learned yesterday and create a model with the classification we want? 02:14:53 Tavis Anderson: best practice?! I like to stick things randomly all over the place. It’s 100% reproducible that way. 02:18:31 Alex Styer (he/him): I’ve got a bunch of images of rice plants against a black background so there’s very little noise around the plant/object I’m interested in and most images are incredibly similar (little variation from plant to plant)—to do some sort of transfer learning or de novo object detection do you have a sense of how many training images would be required? 02:19:55 Alex Styer (he/him): Nooooooooooo :( 02:21:25 Alex Styer (he/him): Cool so there is hope! 02:40:13 Heather Jones: Hello, When everyone returns from lunch, I would appreciate your participation in a short survey about the course. Laura will step away for a few minutes, and I will send the link. Thanks! 03:08:56 Heather Jones: HI, When everyone is back from lunch, please use your emoji tool to let me know you are back 03:11:32 Heather Jones: HI, When everyone is back from lunch, please use your emoji tool to let me know you are back 03:14:28 Heather Jones: https://www.menti.com/hvby56771g 03:14:40 Heather Jones: Please join me briefly to complete a short course eval. 03:23:21 Alex Styer (he/him): Just a general question about resolution—is it best to train networks on low resolution copies of images and then apply it to the full resolution originals? 03:24:50 Alex Styer (he/him): Yep! 03:25:04 Aaron Szczepanek: I went to train the yolo3 model using the method we did yesterday. I found the model was all convolution layers.. no dense layers? 03:26:28 Aaron Szczepanek: this is outputting the box? 03:27:21 Aaron Szczepanek: cool 03:30:45 Laura Boucheron: Mask-RCNN 03:44:47 WGMeikle: My iterations was set at '4000'. I also didn't have the import struct command so maybe I have an old version? 03:50:09 Tavis Anderson: Has anyone else had a “TypeError: Tensors are unhashable.” Error in this area? 03:51:39 Tavis Anderson: That was what google told me too. Thanks. 03:52:08 Aaron Szczepanek: I had to update my tensor flow. I had 1.3.1 which had worked until today 03:52:13 Brian Stucky: Before lunch, I was helping several people who had tensorflow 1.x for some reason. 03:52:36 Brian Stucky: Unclear why conda was not installing a newer version by default. 03:52:53 Tavis Anderson: I have 2.4.1 03:53:47 WGMeikle: Running the "Look at output" block with 100 iterations is taking a long time. Is that a problem? 03:55:32 WGMeikle: You're right 04:02:05 Aaron Szczepanek: what if you put in the other model? 04:02:57 Aaron Szczepanek: like yolo3 04:09:28 Sean Kearney (USDA-ARS): Is there a variation of this style transfer approach to look at what is different between two images (in the 'eyes' of the model)? 04:13:43 Alex Styer (he/him): That’s definitely a tower made of chilis 04:36:14 Sean Kearney (USDA-ARS): Thanks so much Laura - that was a great workshop! 04:36:17 Surya Saha: Is there a reason yesterday’s notebook has not been posted yet? 04:36:19 Aaron Szczepanek: do they use this for audio clips? 04:36:28 Surya Saha: Great Workshop, of course! 04:36:40 Peihua_UMD: Does there are any pubulications applied GAN in agricultural tasks? 04:37:00 Joe Kawash: Fantastic workshop! 04:37:26 Aaron Szczepanek: i thought i saw one that had taken images of the wavelengths 04:37:35 Alex Styer (he/him): Similar to Peihua’s Q—are there networks/models that are more specifically built for plants/crops? 04:37:54 Jonathan Shao: This has been a great workshop. Great detail. Thanks so much. 04:38:36 Peihua_UMD: And another question is if there are any way to embedding image with text data? 04:38:58 Surya Saha: A more generic question - Can you comment on using the tools covered here for non-image data? For e.g. sequence data, etc. 04:39:02 Heather Jones: All notebooks are now visible 04:39:32 Peihua_UMD: since sometimes, our task was not totally with image, but with some wavelength data toegether 04:39:59 Alex Styer (he/him): Very cool! 04:40:51 Jonathan Shao: What is the minimum number of pictures that you would consider appropriate for deep learning. 04:40:56 Alex Styer (he/him): Thank you!! Workshop has been great and really helpful learning some of the jargon for things I’ve been trying to muddle my way through on my own… 04:46:15 Peihua_UMD: https://github.com/brightmart/text_classification 04:46:28 Tavis Anderson: Hie, B., Zhong, E.D., Berger, B. and Bryson, B., 2021. Learning the language of viral evolution and escape. Science, 371(6526), pp.284-288. 04:46:31 Laura Boucheron: Recurrent Neural Network (RNN), Long Short Term Memory (LSTM) 04:47:26 Surya Saha: Thanks! 04:48:09 Surya Saha: Yep, they are 04:48:27 Jonathan Shao: For datasets that have less images (hard to produce images in the lab) is augmenting the data via rotation for example appropriate to boost the number of images where you have less images 04:52:57 Laura Boucheron: NN-SVG 04:54:51 Laura Boucheron: https://github.com/ashishpatel26/Tools-to-Design-or-Visualize-Architecture-of-Neural-Network 04:58:32 Jonathan Shao: thanks 04:58:58 Heather Jones: SIgning off. Have a great weekend!