Shadowing Practice: Controlled Multi-modal Image Generation for Plant Growth Modeling - Learn English Speaking with Video

Ders oluşturuluyor...
1
Hello my name is Libana Roscher
2
and I want to present our paper Controlled multimodal image generation for plant growth modeling
3
which was presented at the international conference on pattern recognition last summer.
4
So when we have a look at such an aquaculture field
5
and we work with robotics and with machine learning methods for phenotyping
6
we can ask ourselves different questions
7
so for example we can ask what is on the field
8
which is a typical classification task
9
or we can ask what is the yield in a specific area
10
which is a typical regression task
11
or we can ask ourselves what will a plant look like in the future
12
and this is a forecasting tasks
13
and this is exactly the question we addressed in our paper
14
so imagine the following scenario you already know that
15
when plants grow closely together this can be very beneficial
16
so for example it can increase the biomass the crucial question is now what is actually a good mixture
17
and so what you can do is you can test different mixtures then you can collect data
18
and then you can learn a model from this information
19
and this model relates the information from what we have measured to a parameter like the biomass
20
so you have a relationship between the mixtures and the biomass
21
and then the good thing is this model can be applied to new mixtures
22
and these new mixtures were not tested yet in the field before
23
so you get an idea what potential good mixtures are
24
And one problem is now that one might be interested in the biomass
25
and the other might be interested in the spread of diseases.
26
So it would be good to combine this and to get complete information.
27
And this can be done if you model the whole appearance with plants, that means if you generate images.
28
And with this you are flexible in what you do with the result
29
because it can be treated such as an artificial sensor measurement.
30
So in order to generate such images we used
31
so -called generative adversarial networks illustrated here and this is a neural network consisting of two parts.
32
So one part is the generator G and this generator is responsible to generate images.
33
The other part is the discriminator, and the discriminator helps to train the generator to generate really good images.
34
The discriminator does it by trying to distinguish between real images and images which were generated by the generator.
35
So in the best case, the generator generates such good images
36
that the discriminator can no longer distinguish between a real image
37
and an image which was generated but there's one crucial thing
38
which is important for our application
39
so we do not want only want to generate an arbitrary
40
nice -looking image we want to condition on specific factors on specific things
41
and one important thing is we want to condition on the appearance of an of an additional input image
42
which is from an earlier point in time
43
and the second important thing is we want to condition on
44
so -called factors of variation and these factors of variation are factors
45
which influence the appearance of the of the plants
46
and the novelty of our paper is exactly this part
47
because on the one hand we learned the representation
48
which is useful for the machine learning model but on the other side it was
49
Because we can also easily and intuitively change these factors of variation so that we can produce different scenarios, different outcomes.
50
And so here you can see results of
51
which we presented in our paper where we generated images with chain factors of variation.
52
So our dataset contains UAV images, which were captures in Kampus -Klein -Altendorf,
53
where we have different mixtures of spring wheat and faba bean.
54
And we also collected different factors of variation, which belong to these mixtures.
55
And they are illustrated in green and light blue and in dark blue.
56
And on the left, you can see the input image, which is from an early growth stage.
57
And this is the real scene from a later growth stage and the measured factors of variation.
58
And these images are all generated images.
59
So here we change the factors of variation.
60
So for example, on the very right image, you can see we set all factors of variation for Faba bean to zero.
61
And actually there are no Faba bean plants in the generated image.
62
So we can clearly control and manipulate how the images can be generated it
63
and from these images we can now derive specific parameters pointing
64
us to potential promising mixtures we can later test in the
65
field with this thank you very much for listening you can find the paper by scanning the qr code
66
and if you are more interested in this research direction you can go to our website
67
or you can have a look in my YouTube channel.
68
Thanks a lot!

Why This Video Is Perfect for Language Learning

This video is a goldmine for speaking practice! The speaker, Libana Roscher, uses clear, conversational English to explain complex ideas—making it ideal for learners to mimic natural rhythm and intonation. Her pace is steady, with pauses that let you follow along, and she uses common phrases like "so for example" and "the crucial question is now" that are easy to incorporate into your own speech. Plus, the topic blends technical terms with everyday language, giving you a chance to practice both formal and casual expression.

Breaking Down Natural Speech Patterns

Let’s look at a few lines that sound effortlessly natural. First: "Hello my name is Libana Roscher and I want to present our paper..." The flow here is smooth—no forced pauses—showing how native speakers connect short sentences in introductions. Next: "So when we have a look at such an aquaculture field and we work with robotics..." The word "so" here is a filler that softens the start of a thought, a common trick to sound more conversational. Finally: "The discriminator does it by trying to distinguish between real images and images which were generated..." Notice how "which were" is slightly reduced in speech—this contraction (or near-contraction) is key to sounding natural, not robotic. These small details make the dialogue relatable and easy to replicate.

A Simple Practice Routine Using the Shadowing Technique

Ready to improve your pronunciation and fluency? Try this shadowing technique loop:

  • Watch a 10-second clip of the video, focusing on the speaker’s mouth movements and intonation.
  • Hit pause, then immediately repeat the clip out loud, mimicking her pace and stress (e.g., emphasizing "crucial" in "the crucial question").
  • Record yourself and compare it to the original. Notice if you’re pausing in the same places or stressing the right words.
  • Repeat with a new clip, gradually increasing to 30 seconds. Do this daily for 10 minutes, and you’ll quickly notice your speech sounding more natural.
This routine turns passive listening into active speaking practice, helping you build muscle memory for real-world conversations. Whether you’re working on pronunciation or fluency, this video and the shadowing technique are powerful tools for quick progress!

Gölgeleme Tekniği Nedir?

Gölgeleme, başlangıçta profesyonel tercüman eğitimi için geliştirilen ve çok dilli Dr. Alexander Arguelles tarafından popüler hale getirilen, bilim destekli bir dil öğrenme tekniğidir. Yöntem basit ama güçlüdür: ana dili İngilizce olan bir sesi dinler ve hemen yüksek sesle tekrar edersiniz — konuşmacıyı 1-2 saniye gecikmeyle takip eden bir gölge gibi. Pasif dinleme veya dilbilgisi alıştırmalarının aksine, gölgeleme beyninizi ve ağız kaslarınızı gerçek konuşma kalıplarını eşzamanlı olarak işlemeye ve yeniden üretmeye zorlar. Araştırmalar, telaffuz doğruluğu, tonlama, ritim, bağlı konuşma, dinleme anlama ve konuşma akıcılığını önemli ölçüde geliştirdiğini göstermektedir — bu da onu IELTS Konuşma hazırlığı ve gerçek dünya İngilizce iletişimi için en etkili yöntemlerden biri yapar.