Calculating Convolution sizes is something that I found particularly hard after understanding convolutions for the first time.

I couldn't remember the formula because I didn't understand its working exactly.

So here's my attempt to get some intuition behind the calculation.🔣👇

BTW if you haven't read the thread 🧵 on 1D, 2D, 3D CNN, you may want to check it out once.

https://t.co/kxZ6ZeHCrk
First, observe the picture below🖼
The 2 x 2 filter slides over the
3 rows, 2 times and,
4 columns, 3 times

So, let's try subtracting the filter size first
3 - 2 = 1
4 - 2 = 2

Looks short, we'll need to compensate the 1 in both.
3 - 2 + 1 = 2
4 - 2 + 1 = 3

hence the formula so far becomes:
Now let's discuss padding0⃣

Zero padding makes it possible to get output equal to the input by adding extra columns.

It provides extra space for the sliding, making up for the lost space
Padding of p would mean increasing the input size by adding p to both sides

Considering width, there will be padding for left and for right, both equal, same for the height.

The modified formula becomes
All of our calculation so far assumes that we are taking one step at a time during sliding, a stride of 1

What if we take more than that?🏃

We will be cutting our distance short by increasing the size of our leap. So to account for this we will divide with stride size.
Keep in mind that we should make sure that the calculation doesn't go into decimals.

We generally select our values in such a way that the calculations result in an integer.
Now as we may remember from the last thread, that one filter leads to 1 output, be it 1D, 2D...

So the depth of the output will be equal to the number of filters applied.
With all that in mind, let's try to solve a simple question below:
We can try the same using Keras and its functions.
This website is a ConvNet shape calculator with which you can play around a little bit for better understanding.

https://t.co/8rU5BOAkts
Now if you feel like you can calculate correctly, try to pick any network and calculate its output sizes, validate using the model summary.

Or pick 3D convolutions and calculate its outputs, the principle remains the same.
All the above points helped me to be able to understand the CNN architectures better and not be bothered by the output summary.

Hope this helps you too! 👍

More from All

MASTER THREAD on Short Strangles.

Curated the best tweets from the best traders who are exceptional at managing strangles.

• Positional Strangles
• Intraday Strangles
• Position Sizing
• How to do Adjustments
• Plenty of Examples
• When to avoid
• Exit Criteria

How to sell Strangles in weekly expiry as explained by boss himself. @Mitesh_Engr

• When to sell
• How to do Adjustments
• Exit


Beautiful explanation on positional option selling by @Mitesh_Engr
Sir on how to sell low premium strangles yourself without paying anyone. This is a free mini course in


1st Live example of managing a strangle by Mitesh Sir. @Mitesh_Engr

• Sold Strangles 20% cap used
• Added 20% cap more when in profit
• Booked profitable leg and rolled up
• Kept rolling up profitable leg
• Booked loss in calls
• Sold only


2nd example by @Mitesh_Engr Sir on converting a directional trade into strangles. Option Sellers can use this for consistent profit.

• Identified a reversal and sold puts

• Puts decayed a lot

• When achieved 2% profit through puts then sold

You May Also Like

First update to https://t.co/lDdqjtKTZL since the challenge ended – Medium links!! Go add your Medium profile now 👀📝 (thanks @diannamallen for the suggestion 😁)


Just added Telegram links to
https://t.co/lDdqjtKTZL too! Now you can provide a nice easy way for people to message you :)


Less than 1 hour since I started adding stuff to https://t.co/lDdqjtKTZL again, and profile pages are now responsive!!! 🥳 Check it out -> https://t.co/fVkEL4fu0L


Accounts page is now also responsive!! 📱✨


💪 I managed to make the whole site responsive in about an hour. On my roadmap I had it down as 4-5 hours!!! 🤘🤠🤘