I am using a Swin Transformer for a hierarchical problem of multi calss multi label classification. I would like to visualize the self attention maps on my input image trying to extract them from the model, unfortunately I am not succeeding in this task. Could you give me a hint on how to do it?
I share you the part of the code in which I am trying to do this task.
attention_maps = []
for module in model.modules():
#print(module)
if hasattr(module,'attention_patches'): #controlla se la variabile ha l' attributo
print(module.attention_patches.shape)
if module.attention_patches.numel() == 224*224:
attention_maps.append(module.attention_patches)
for attention_map in attention_maps:
attention_map = attention_map.reshape(224, 224, 1)
plt.imshow(sample['image'].permute(1, 2, 0), interpolation='nearest')
plt.imshow(attention_map, alpha=0.7, cmap=plt.cm.Greys)
plt.show()
``
In addition if you know about some explainability techniques, like Grad-CAM, which could be used with a hierarchical Swin Transformer, feel free to attach a link, it would be very helpful for me.
I am also researching the same, while I don't have anything specific to SWIN. Here are some resources related to Vision Transformers. I hope it helps:
https://jacobgil.github.io/deeplearning/vision-transformer-explainability
https://github.com/jacobgil/vit-explain
https://github.com/hila-chefer/Transformer-Explainability
https://github.com/hila-chefer/Transformer-Explainability/blob/main/Transformer_explainability.ipynb
Related
I want to display data from Parse in a list from GamesScores class using Container in Codename One, this is what I've tried so far and it's not showing anything nor giving any errors:
Container container = findListCont();
container.setLayout(BoxLayout.y());
container.setScrollableY(true);
ParseQuery<ParseObject> query = ParseQuery.getQuery("GameScore");
List<ParseObject> results = (List<ParseObject>) query.find();
System.out.println("Size: " + results.size());
container.addComponent(results, f);
Please help me out, I'm a new in Codename One. If there tutorials on it, please share or anything to help me achieve the desired results.
I'm actually shocked this isn't failing. You are using the add constraint to place the object result as a constraint and you add the form object into the container...
You need to loop over the results and convert them to components to add into the layout. It also seems that you are using the old GUI builder which I would recommend against.
Generally something like this rough pseudo code should work assuming you are using a box Y layout:
for(ParseObject o : results) {
MultiButton mb = new MultiButton(o.getDisplayValue());
f.add(mb);
}
f.revalidate();
does anyone use arima.rob() function described by Eric Zivot and Jiahui Wang in { Modelling Financial Time Series with S-PLUS } ?
I have a question about it:
I used a dataset of network traffic flows that has anomaly, and I tried to predict the last part of dataset by robust ARIMA method (Arima.rob() function) .I compare this model with arima.mle of S-PLUS. But Unexpectedly, arima.rob’s prediction did not better than that.
I’m not sure my codes are correct and may be the reason of fault is my codes.
Please, help me if I used Arima.rob inappropriately?
tmp.rr<-arima.rob((tmh75)~1,p=2,d=1,q=2,freq=24,maxiter=4,max.fcal=80000)
tmp.for<-predict(tmp.rr,n.predict=10,newdata=df1,se=T)
plot(tmp.for,tmh75)
summary(tmp.for)
my code for classic arima:
`model <- list(list(order=c(2,1,2)),list(order=c(3,1,2),period=24))
fith <- arima.mle(tmh75-mean(tmh75),model=model)
foreh <- arima.forecast(tmh75,n=25,model=fith$model)
tsplot(tmh75,foreh$mean,foreh$mean+foreh$std.err,foreh$mean-foreh$std.err)
`
I am using Maya to do some procedural work, and I have a lot of textures that I need to load into Maya, and they all have transparencies (alpha channels). I would very much like to be able to automate this process. Using PyMEL, I can create my textures and hook them up to a shader, but the alpha doesn't set properly by default. There is an attribute in the psdFileTex node called "Alpha to Use", and it must be set to "Transparency" in order for my alpha channel to work. My question is this - how do I use PyMEL scripting to set the "Alpha to Use" attribute properly?
Here is the code I am using to set up my textures:
import pymel.core as pm
pm.shadingNode('lambert', asShader=True, name='myShader1')
pm.sets(renderable=True, noSurfaceShader=True, empty=True, name='myShader1SG')
pm.connectAttr('myShader1.outColor', 'myShader1SG.surfaceShader', f=True)
pm.shadingNode('psdFileTex', asTexture=True, name='myShader1PSD')
pm.connectAttr('myShader1PSD.outColor', 'myShader1.color')
pm.connectAttr('myShader1PSD.outTransparency', 'myShader1.transparency')
pm.setAttr('myShader1ColorPSD.fileTextureName', '<pathway>/myShader1_texture.psd', type='string')
If anyone can help me, I would really appreciate it.
Thanks
With any node, you can use listAttr() to get the available editable attributes. Run listAttr('myShaderPSD'), note in it's output, there will be two attributes called 'alpha' and 'alphaList'. Alpha, will return you the current selected alpha channel. AlphaList will return you however many alpha channels you have in your psd.
Example
pm.PyNode('myShader1PSD').alphaList.get()
# Result: [u'Alpha 1', u'Alpha 2'] #
If you know you'll only ever be using just the one alpha, or the first alpha channel, you can simply do this.
psdShader = pm.PyNode('myShader1PSD')
alphaList = psdShader.alphaList.get()
if (len(alphaList) > 0):
psdShader.alpha.set(alphaList[0])
else:
// No alpha channel
pass
Remember that lists start iterating from 0, so our first alpha channel will be located at position 0.
Additionally and unrelated, while you're still using derivative commands of the maya.core converted for Pymel, there's still some commands you can use to help make your code read nicer.
pm.setAttr('myShader1ColorPSD.fileTextureName', '<pathway>/myShader1_texture.psd', type='string')
We can convert this to pymel like so:
pm.PyNode('myShader1ColorPSD').fileTextureName.set('<pathway>/myShader1_texture.psd')
And:
pm.connectAttr('myShader1PSD.outColor', 'myShader1.color')
Can be converted to:
pm.connect('myShader1PSD.outColor', 'myShader1.color')
While they may only be small changes, it reads just the little bit nicer, and it's native PyMel.
Anyway, I hope I have helped you!
I state that Parse.com use for my app
Does anyone know what 's the right way to create a list of friends?
The user must be able to create a relationship with another user.
I tried to take a look at AnyPic but I could not follow. seems very complicated for me ... I know there is the possibility of creating a relationship with PFRelation but did not find much nl web
Can you help?
Thank's Rory
If you want to have both sides be able to see the friendship you have two options:
duplicate the relationship, i.e. a PFRelation on each PFUser
us a many-to-many table, i.e. a new Class with two PFUser references, and possibly other information
Given that you might want more information about the relationship (e.g. status=requested/accepted/rejected, etc), I would suggest option two.
Here's a similar question on managing friend requests and friend lists using Parse.
then, I had to make a query where I drew all the posts of my users (Timeline) and their names ... I had a problem (same as what I here to do this' you suggested) or call in a query of the Pointer ...
I solved so
This is the content of the cell
NSString *user = [[object objectForKey:#"Utente"] valueForKey:#"Nome_Cognome"];
cell.FFNomeLabel.frame=CGRectMake(15, -35, 270, 100);
cell.FFNomeLabel.textAlignment = NSTextAlignmentRight;
cell.FFNomeLabel.text = user;
[cell.BiancaView addSubview:cell.FFNomeLabel];
NSString *img = [[object objectForKey:#"Utente"] valueForKey:#"foto"];
cell.FFImmagineUtente.file = (PFFile *)img;
cell.FFImmagineUtente.frame = CGRectMake(10, 10, 70, 70);
[cell.FFImmagineUtente.layer setMasksToBounds:YES];
[cell.FFImmagineUtente.layer setCornerRadius:35.3f];
cell.FFImmagineUtente.contentMode = UIViewContentModeScaleAspectFill;
[cell.FFImmagineUtente loadInBackground];
What do you think?
Also how can I save a pointer that is not a Current User?
I saw the documentation parse, but having little experience I was not entirely clear :)
I wish to search twitter for a word (let's say #google), and then be able to generate a tag cloud of the words used in twitts, but according to dates (for example, having a moving window of an hour, that moves by 10 minutes each time, and shows me how different words gotten more often used throughout the day).
I would appreciate any help on how to go about doing this regarding: resources for the information, code for the programming (R is the only language I am apt in using) and ideas on visualization. Questions:
How do I get the information?
In R, I found that the twitteR package has the searchTwitter command. But I don't know how big an "n" I can get from it. Also, It doesn't return the dates in which the twitt originated from.
I see here that I could get until 1500 twitts, but this requires me to do the parsing manually (which leads me to step 2). Also, for my purposes, I would need tens of thousands of twitts. Is it even possible to get them in retrospect?? (for example, asking older posts each time through the API URL ?) If not, there is the more general question of how to create a personal storage of twitts on your home computer? (a question which might be better left to another SO thread - although any insights from people here would be very interesting for me to read)
How to parse the information (in R)? I know that R has functions that could help from the rcurl and twitteR packages. But I don't know which, or how to use them. Any suggestions would be of help.
How to analyse? how to remove all the "not interesting" words? I found that the "tm" package in R has this example:
reuters <- tm_map(reuters, removeWords, stopwords("english"))
Would this do the trick? I should I do something else/more ?
Also, I imagine I would like to do that after cutting my dataset according to time (which will require some posix-like functions (which I am not exactly sure which would be needed here, or how to use it).
And lastly, there is the question of visualization. How do I create a tag cloud of the words? I found a solution for this here, any other suggestion/recommendations?
I believe I am asking a huge question here but I tried to break it to as many straightforward questions as possible. Any help will be welcomed!
Best,
Tal
Word/Tag cloud in R using "snippets" package
www.wordle.net
Using openNLP package you could pos-tag the tweets(pos=Part of speech) and then extract just the nouns, verbs or adjectives for visualization in a wordcloud.
Maybe you can query twitter and use the current system-time as a time-stamp, write to a local database and query again in increments of x secs/mins, etc.
There is historical data available at http://www.readwriteweb.com/archives/twitter_data_dump_infochimp_puts_1b_connections_up.php and http://www.wired.com/epicenter/2010/04/loc-google-twitter/
As for the plotting piece: I did a word cloud here: http://trends.techcrunch.com/2009/09/25/describe-yourself-in-3-or-4-words/ using the snippets package, my code is in there. I manually pulled out certain words. Check it out and let me know if you have more specific questions.
I note that this is an old question, and there are several solutions available via web search, but here's one answer (via http://blog.ouseful.info/2012/02/15/generating-twitter-wordclouds-in-r-prompted-by-an-open-learning-blogpost/):
require(twitteR)
searchTerm='#dev8d'
#Grab the tweets
rdmTweets <- searchTwitter(searchTerm, n=500)
#Use a handy helper function to put the tweets into a dataframe
tw.df=twListToDF(rdmTweets)
##Note: there are some handy, basic Twitter related functions here:
##https://github.com/matteoredaelli/twitter-r-utils
#For example:
RemoveAtPeople <- function(tweet) {
gsub("#\\w+", "", tweet)
}
#Then for example, remove #d names
tweets <- as.vector(sapply(tw.df$text, RemoveAtPeople))
##Wordcloud - scripts available from various sources; I used:
#http://rdatamining.wordpress.com/2011/11/09/using-text-mining-to-find-out-what-rdatamining-tweets-are-about/
#Call with eg: tw.c=generateCorpus(tw.df$text)
generateCorpus= function(df,my.stopwords=c()){
#Install the textmining library
require(tm)
#The following is cribbed and seems to do what it says on the can
tw.corpus= Corpus(VectorSource(df))
# remove punctuation
tw.corpus = tm_map(tw.corpus, removePunctuation)
#normalise case
tw.corpus = tm_map(tw.corpus, tolower)
# remove stopwords
tw.corpus = tm_map(tw.corpus, removeWords, stopwords('english'))
tw.corpus = tm_map(tw.corpus, removeWords, my.stopwords)
tw.corpus
}
wordcloud.generate=function(corpus,min.freq=3){
require(wordcloud)
doc.m = TermDocumentMatrix(corpus, control = list(minWordLength = 1))
dm = as.matrix(doc.m)
# calculate the frequency of words
v = sort(rowSums(dm), decreasing=TRUE)
d = data.frame(word=names(v), freq=v)
#Generate the wordcloud
wc=wordcloud(d$word, d$freq, min.freq=min.freq)
wc
}
print(wordcloud.generate(generateCorpus(tweets,'dev8d'),7))
##Generate an image file of the wordcloud
png('test.png', width=600,height=600)
wordcloud.generate(generateCorpus(tweets,'dev8d'),7)
dev.off()
#We could make it even easier if we hide away the tweet grabbing code. eg:
tweets.grabber=function(searchTerm,num=500){
require(twitteR)
rdmTweets = searchTwitter(searchTerm, n=num)
tw.df=twListToDF(rdmTweets)
as.vector(sapply(tw.df$text, RemoveAtPeople))
}
#Then we could do something like:
tweets=tweets.grabber('ukgc12')
wordcloud.generate(generateCorpus(tweets),3)
I would like to answer your question in making big word cloud.
What I did is
Use s0.tweet <- searchTwitter(KEYWORD,n=1500) for 7 days or more, such as THIS.
Combine them by this command :
rdmTweets = c(s0.tweet,s1.tweet,s2.tweet,s3.tweet,s4.tweet,s5.tweet,s6.tweet,s7.tweet)
The result:
This Square Cloud consists of about 9000 tweets.
Source: People voice about Lynas Malaysia through Twitter Analysis with R CloudStat
Hope it help!