TensorFlow Reviews

I am using TensorFlow for many computing projects, such as image data analysis, data compressions, and data processing.

I have used TensorFlow GS, TensorFlow for mobile, and TensorFlow Lite. The way things operate is changing from a server-side to edge computing and mobile devices. I have been working on how to reduce the resource usage, we have downgraded and should have good performance with the best accuracy in our machine and models.

Edge computing has some limited resources but TensorFlow has been improving in its features. It is a great tool for developers.

There are a lot of problems, such as integrating our custom code. In my experience model tuning has been a bit difficult to edit and tune the graph model for best performance. We have to go into the model but we do not have a model viewer for quick access.

There should be better integration and standardization with different operating systems. We need to always convert from one model to another and there is not a single standardized model output that we could use on different platforms, such as Intel x56, x64 based, AR-based, or Apple M1 chips.

I have been using this solution within the last 12 months.

TensorFlow is a reliable solution, but we have not explored all the aspects of TensorFlow. We have been building our customized applications, such as libraries, features, or functions. We only use the features that allow our application to work. Different areas need to be researched on a low level to make them more efficient.

We have been working on specific applications and any model built on TensorFlow can be applied to any scalable level. The solution has built-in scalability.

I have not needed to use technical support.

I did not require a license for this solution. It a free open-source solution.

I rate TensorFlow a ten out of ten.

I primarily used the solution for computer vision applications, for example, detection and segmentation, and OCR.

We used an architecture from a published paper. It was based on TensorFlow and we upgraded it and developed on it. I also worked on face verification and likeness detection. We are working on anti-spoofing detection. We did some things around face verification and likeness detection. I used TensorFlow specifically.

I've also used the solution to detect hands, tracking customers in the supermarkets, and using the solution for detecting the pickup and dropping of objects from shelf to basket in the supermarket.

Lately, I've been working on a project in Arabic that is designed to detect handwriting. I also use PyTorch to help with this task.

For one particular project, we did an extraction of the Arabic language from a crucial document, like an ID. We needed to capture the ID using the application so that the application sends the ID to the server. We needed to make an Egyptian ID detection on mobile. I built a simple commercial network to customize the ID and converted it into TensorFlow White and made some compositions to make it faster to run. We deployed it on the mobile. For this bot, there's full support in this area, which is great.

The solution is quite useful for production. It tends to provide for digital devices or mobile devices. You can deploy your model on Android or iOS. I did that before on Android. It provides TensorFlow GS or JavaScript to run TensorFlow applications in the browser. 

It's quite a valuable solution when we go to production.

Google is behind TensorFlow, and they provide excellent documentation. It's very thorough and very helpful. 

Overall, the solution has been quite helpful. I can't recall missing any features when I was using it.

I know this is out of the scope of TensorFlow, however, every time I've sent a request, I had to renew the model into RAM and they didn't make that prediction or inference. This makes the point for the request that much longer. If they could provide anything to help in this part, it will be very great.

The solution is quite stable. It's reliable. It doesn't crash or freeze. There are no bugs or glitches to deal with. We've been happy with the performance.

The scalability of the solution is good. When it comes to TensorFlow (or PyTorch) you can train on multiple GPUs at the same time, and multiple machines at the same time, also. 

In TensorFlow, I didn't train on multiple GPUs, yet, however, I know it's very easy and straightforward. I've been through it. Scaling should be a problem for a company. If they want to scale, they can.

I've never been in touch with technical support. I can't speak to their level of knowledge or how quickly they respond.

I also use PyTorch and Amazon SageMaker.

Amazon Solution provides the how-to where we can use PyTorch and TensorFlow to train models on huge datasets on AWS end-user. They are complementing each other on a project.

The entire implementation process is quite straightforward. It's not complex at all.

The deployment is very fast. You can have it up and running in five minutes. You just go and install the version you would like to use and you are done. It's all very simple.

I always go with the deployment with TensorFlow. I don't try to use PyTorch in this area. A year ago, when I was deploying a semantic segmentation model on the server, each time I sent a request that I need to reload the model input that ends.

It's my understanding that the version we use is free. It doesn't cost any money.

When we did the utilization applications, we were deploying on digital ocean servers. For the projects that I'm working on now, we are planning to deploy it on its own port attached to the robot. We haven't done it, yet. We are finishing the project right now. For deploying the solutions, I deploy them on the digital ocean. 

I'd recommend the solution. I'd also recommend users considering the solution do a bit of studying. There are some great courses on Coursera and there's a recent one called DeepLearning.AI that is extremely useful.

Overall, as I use the product pretty much everywhere, I would rate it at a ten out of ten.

I use this solution to create Neural Networks, which are computer algorithms for the recognition of objects. This is done based on the SL object that predefines it.

Most of our experience is computer related, but in most cases, we work with images.

Our company was working on a specific project that required a graph. It is now under development and TensorFlow allows us to implement this functionality for the customer who needs to work with recognizing and defining the special mark on the student's workbooks.

The most valuable features are the frameworks and the functionality to work with different data, even when we have a certain quantity of data flowing.

There are connection issues that interrupt the download needed for the data sets. We need to prepare them ourselves.

I have been using TensorFlow for one year.

I have experience not just in TensorFlow, but in the TensorFlow Keras, beginning from TensorFlow 2.0, there are package Keras in TensorFlow. Using this cache, I have created some Neural Networks on Python.

I am using the latest version.

It's a stable product.

It's a scalable solution and we can scale it for different tasks.

We have two specialists that are connected to TensorFlow.

We have not contacted technical support.

Before using TensorFlow, we used different neural networks that were based on Darknet.

The initial setup was easy. There is a lot of documentation available and it was not a problem for us.

It was easy to install.

Setting up TensorFlow on the local computer will take one to two hours to complete. However, if it is for an industrial product that has entered the market and needs to work in the real environment, it would depend on the goal and task that we are working on. 

We completed the installation ourselves without any external help.

Maintenance is based on the customer's needs. We have approximately 40 developers, so if the customer requires maintenance and support then we can provide that for them.

We are using the free version.

I would recommend TensorFlow for techniques that need to develop Neural Networks. I would also recommend PyTorch.

I would rate this solution a nine out of ten.

We have a project that a Canada-based client is expecting us to develop. If there is a hardware product, it's a mirror LCD device, that is installed in your home and when you start doing an exercise, our AI algorithm will detect what kind of exercise, whether you're doing pushups, jump, etc. We also detect what kind of hardware equipment is being used. We also use TensorFlow to count. 

TensorFlow improves my organization because our clients get a lot of investment from their investors and we progressively improve the products. Every six months we release new features. 

They have a very good vision and roadmap for the next two years. 

TensorFlow is like a library. PyTorch is also a library. These are deep learning libraries that provide a set of functions. Ultimately you have to build a framework. TensorFlow as a whole is useful to us because we use a lot of functions, like activation functions or volition functions, feature mapping, and feature extraction. 

In terms of improvement, we always look for ways they can optimize the model, accelerate the speed and the accuracy, and how can we optimize with our different techniques. There are various techniques available in TensorFlow. Maintaining accuracy is an area they should work on. When there are more and more objects involved with the model, the models get confused. So maintaining the accuracy and speed with the number of classes is the biggest area for improvement. It is a major challenge that we are seeing right now and we are trying to solve the problem.

It's quite stable. It also helps while we have to take it to the browser platform as well as the Android platform because Google is developing in such a way that we can easily migrate. I'll definitely go with TensorFlow when we have to deal with the different platforms. We can easily convert into TF Lite. We can run the model into the browser, as well as on the Android platform.

We have 13 developers and 10 developers mainly focus on TensorFlow, PyTorch, and all the deep learning things. The rest of the developers are C+ and computer vision developers.

If any maintenance is required our team is capable.

Scalability is a major challenge because under the exercise project right now, we have integrated 14 exercises, one port and client is targeting more than 50 exercises. That's where we need scalability. 

Sometimes when we face some issues, we mostly get solutions from stakeholders. So we are not using that much technical support from the TensorFlow team.

TensorFlow is from Google, PyTorch is from Facebook. PyTorch is mostly compatible with Python. If you have to consider it in different platforms like different languages C+, Android, then TensorFlow is really good. In that sense, I'd go with TensorFlow, but PyTorch is really stable. It's only helpful when we are dealing with the Python language. That's where it's really helpful. So both have some advantages and disadvantages.

We got some help from the internet blogs but by now our team is really capable. If there are any issues or errors with a particular version, they can immediately deploy changes. It's now a completely smooth process.

We use the open-source version.

There are always new versions coming out and some versions have issues while some versions don't. When you deploy with the latest version, just make sure that all the systems work as expected when you're deploying. 

I would rate TensorFlow an eight out of ten. 

The main purpose of TensorFlow is to  develop neural networks for data science projects. For example, I had a project about a super-resolution GAN, which is a model that you give a low-resolution image, and it will complete the details for you. I used Keras and TensorFlow for this model and it was really easy to use. The time to implement was simply minimal in comparison to the time for testing, logic, and high-level implementation. That was the highlight of my academic project. For a client, I used TensorFlow and Keras to develop a predictive heat map for orders. He wanted to build a predictive model for a taxi company. They wanted to tell the drivers, "Okay, this area has more probability of having higher orders than another area." I used TensorFlow and Keras to develop a model to predict the areas which have a higher probability and built a heat map to show the drivers. That is actually the highlight of my industrial project. It was a client on Upwork.

For me, it's the simplicity of using the built-in layers. The load from generator method saves us a lot of time and memory in terms of development and the learning process of models, especially networks that have huge parameters. They will need a hundred terabytes of runs but with this method, which is amazing. All the methods of TensorFlow are consistent with each other. So it will save the developer a lot of time and actually, that reflects on the client as well. So the client would get a  high-quality product with a minimal budget. This enables clients with limited budgets to actually hire developers that can develop high-quality models using TensorFlow.

The most valuable feature for TensorFlow is the ability to use CoLab. It's actually also using Torch, but in TensorFlow according to my experience, it's much, much easier to do than the integration with Google CoLab. It's pretty simple to use Google CoLab Pro and use TensorFlow models. It's not a feature, but the best thing about TensorFlow and Keras is that it is the most common in the world and they have huge communities. Whatever error you have, you can actually Google that error and you can get it done in five minutes. So that is, I think, really unique about TensorFlow. I never actually thought about developing a system like TensorFlow. It's so huge and it needs a lot of developers to maintain, but if I want to develop a sub-system that actually helps me to solve a task, I can do that in just two days to develop benchmark models in TensorFlow. If I had to develop this from scratch I would probably need 20 days to a month to develop it myself from scratch. It is also totally Open-Source and free. Open-source applications are not good usually. but TensorFlow actually changed my view about it and I thought, "Look, Oh my God. This is an open-source application and it's as good as it could be." I learned that TensorFlow, by sharing their own knowledge and their own platform with other developers, improved the lives of many people around the globe. If it was a licensed product, a lot of people, for example, in the Middle East or the third world countries, would not be able to help their own communities because of a substantial license fee they cannot afford. The biggest lesson I learned is to have an open-source platform that could impact the world and make it a better place. You get that with TensorFlow.

If I want to develop my own gradient descent, and I want to use the TensorFlow grading descent, but implement it in my own way, it can be difficult. However, if I want to change just one thing in the implementation of TensorFlow functions I have to copy everything that they wrote and change it manually if indeed it can be amended. This is really hard as it's written in C++ and has a lot of complications. But this feature, allowing you to write bespoke code to an implementation of TensorFlow would be really great. Another thing I think that TensorFlow would be much more optimized is to have better CPU versions. I know the problem with Python in general, it lets you only use one thread in the CPU. But even while using TensorFlow, it uses two threads. For example, if I have a high powered CPU, I cannot use it. For example with my laptop, I have a high-powered CPU and I'm using Ubuntu, but my GPU is not recognized. So I can use the CPU, but it's not really optimized for this purpose. Huge calculations require GPU's. I think that could be the second thing that could be optimized. I think TensorFlow 2 has huge improvements over TensorFlow 1. However, it would be really nice if we can actually somehow use the code written in TensorFlow 1, to incorporate it into TensorFlow 2. It generates a lot of errors and you have to change a lot of code and settings. What we can optimize is to actually have consistency between the versions. So TensorFlow 2 is actually a different product, to TensorFlow 1.

I have been using TensorFlow for around one year and a half. I use it for projects. So it has been three years since I took a TensorFlow tutorial or course about TensorFlow. I started applying it in industrial projects and academic projects for one year and a half.

For TensorFlow 2, I think it's very stable. For that version. Stability is inconsistent between the versions of TensorFlow 1 and TensorFlow 2. I think from the TensorFlow functionality side, you don't really need any maintenance, but from the developer's side, you do need maintenance. In my experience, I deployed three or four projects and they just worked consistently. I never had any problems except the problem that I had on my own, bespoke code, but not the code that was actually provided in the TensorFlow library.

I have never used it on multiple servers, as usually I have only dedicated servers, and tend to deploy on on those. I know that it could be scalable to multiple servers like AWS Lambda. I'm not sure, but it can be, although I have never worked with it.

I once posted a question in the questions section on the TensorFlow website and I got an answer in two days and it was really helpful. That actually helped me solve the problem that I had. Apart from that, I didn't really use it because generally, I can find my answer by searching for the error on Google search. It has very good documentation and community support.

I only worked with PyTorch and TensorFlow. I used PyTorch two and a half years ago. I started working with PyTorch and it was difficult for a young person who simply just wants to run models quickly. So in PyTorch, one has to have a huge experience until they deployed their first model. This is really demotivating for someone who is starting to learn the principles. That's a negative for PyTorch and it's a plus for TensorFlow. There is this simplistic structure of TensorFlow such that you can write a fully connected layer model in three lines, and run it on another data set in three or four lines. That is really nice in comparison to PyTorch as you have to write three different classes and data loaders. It's more time consuming to run in PyTorch. So I think it's another positivity for TensorFlow. I can't think of anything you can achieve with PyTorch that you cannot with TensorFlow. That's why I actually just stopped using PyTorch after awhile.

It is very easy to set up as technically it only uses one or two lines if you are using Conda. It is easy to build a simple neural network, by just following the tutorial three lines and that's it. We have running neural networks. Pretty easy. You can host it on your own computer. It's a Python pip package. You can host it on your computer, on a server. The client that I worked with for the heat map prediction, was hosted on a server. It's a website. So the server for calculations was hosted on Amazon web services. So it could be all of these.

I think for learners to deploy a project, you can actually use TensorFlow for free. It's just amazing to have an open-source platform like TensorFlow to deploy your own project. Here in Russia no one really cares about licenses, as it is totally open source and free. My clients in the United States were also pleased to learn when they enquired, that licensing is free. 

I had a problem with it during one implementation. I assumed that the data would be small. I think before implementing your TensorFlow model, it's crucial to note what is the size of your data and will it increase in the future? Usually, a developer wants to develop the model as easily as can it be. So they just tend to load all the data in memory and then run it into a flow model. So that is really problematic if your data is huge. That's why it's best for the developer before they write any line of code to check the data. If it doesn't fit in memory, they can use the TensorFlow functionality of load from a generator and this way they can actually have just one image in the memory at a time per thread. So it's really amazing. So I think that that is the tweak that I would advise developers to have, before developing their model.

I would rate TensorFlow 9 out of 10.

I worked for a French company. They used TensorFlow for image classification. after that, I started working with a Russian-American Company who used TensorFlow mainly for object detection. TensorFlow is very good at object detection. We also used it once for natural language processing and audio processing, but I was not directly involved in that project. I was just assisting with deployment issues. We have some clients which wanted us to deploy on the cloud. Alternatively, some clients are releasing Tenserflow on some new edge devices, as an alternative to deploying on the cloud. It is going to be called NextGen AI or something like that from AWS. We use it for all aspects. including data processing, training, and sometimes deployment, but sometimes the use cases differ in practice for ML. As a result, we sometimes stay with TensorFlow or move into AWS specific architectures.

TensorFlow has benefitted us by enabling faster training and deployment. With TensorFlow, we don't really need any more DevOps to do the deployments. Even data scientists can do the deployment part. This has saved about 30% of the time we used to take for deployments.

Our clients were not aware they were using TensorFlow, so that aspect was transparent. I think we personally chose TensorFlow because it provided us with more of the end-to-end package that you can use for all the steps regarding billing and our models. So basically data processing, training the model, evaluating the model, updating the model, deploying the model and all of these steps without having to change to a new environment. Especially the part where you could train the model again, then evaluate it if it's better than the previous versions. It will do the deployment on its own. The end-users will not really see the change, as the update takes place without any downtime.

It doesn't allow for fast the proto-typing. So usually when we do proto-typing we will start with PyTorch and then once we have a good model that we trust, we convert it into TensorFlow. So definitely, TensorFlow is not very flexible.

We have been using Tensorflow since 2017, so three years.

It's very stable. So we usually don't get any problems. Once any bugs are fixed, you shouldn't have any problems with TensorFlow. Once the deployment process is completed, you can monitor your model and datasets. You can monitor your model to ensure it is correctly deployed and it's working as it's supposed to do, including services.

TensorFlow is very scalable.

I've never really contacted TensorFlow support, but definitely, I can say you don't really need to do that because the support, like the community is pretty strong. Whatever problem you face, there's always going to be some Stack Overflow answer for it or at least some GitHub issue where you can find your solution.

We used PyTorch and MXNet. A couple of my friends actually used MXNet, but I did not use it personally. Now I work mainly between PyTorch and TenserFlow.

Setup got easier with TensorFlow 2. With TensorFlow 1 it was a bit more complicated. There are fewer compatibility issues with the newer version. Training takes at least a week. Deployment usually takes one or two days. With regards to deployment, this changes depending on the client. The usual method is to get our TensorFlow models up and running and then we have to convert them into specific formats depending on the client's requirements. Some clients actually require AWS specific formats. To incorporate that we usually just convert our TensorFlow models to AWS compatible models.

We use in-house teams. I think the ML team has around 20 people. There is a team in Russia, Ukraine and the U.S. I am part of the team in Russia. In Russia, we have around 30 people who have used TensorFlow, including data analysts. They basically handle data pre-processing. We also have ML engineers and ML Ops.

TensorFlow is free, so cost is not an issue.

PyTorch is very flexible, and definitely more flexible than TensorFlow. However, TensorFlow allowed us to deploy our models faster and more robustly. With TensorFlow, you don't encounter a lot of errors. PyTorch is faster to prototype with. So it's very flexible. You can do all the changes that you want, but it's not very stable, whereas TensorFlow is more of a stable solution but is not very flexible. We looked at other alternatives, but they don't scale to large problems.

Have a look at TensorFlow extended. It's very useful. Especially if you know how to use the old system. It will speed up the process of deploying your model. Don't reinvent the wheel. There's always going to be a good GitHub repo out there which kind of answers your solution. You shouldn't really spend a lot of time trying to build the new models where there is some other open source project that actually did a good job of the modelling part. You definitely need to have your own pipelines for this process. Try to build the pipelines that automate most of the tasks for you. Then all you need to concern yourself with is just the architecture. Obtain a pipeline template from GitHub of what you are trying to achieve, amend it for your needs and then you are ready to go. Your model is training already. I would rate TensorFlow 8 out of 10.

I have experience in NRP and time series forecasting and also in marketing-relevant tasks.

For example, I've used the workaround cutoffs to create a deep learning network to classify binary classification. I've done binary classification tasks and multi-label classification tasks. The multi-class classification is based on Hungarian and English texts. I have an ongoing project, where I created an LSTM and this LSTM is able to classify the text for cryptocurrencies. 

Pre-trained models are very important in terms of flow. It's a great opportunity to create very fast, new, deep learning models. 

Before version 2.X, PyTorch had features that were better than this product. Now that it's been updated, it's got all of those missing features and is much better. There's a significant difference.

Users are able to create deployments with Docker and TensorFlow. TensorFlow has a pre-trained model hub. It's a huge hub in a typical NLP or computer vision.

I've used TensorFlow in different areas within marketing tasks. For example, dynamic pricing solutions or classifications as to who will buy something or who will not buy something, or who will return. It's great to use in stock market scenarios, cryptocurrencies, foreign exchange markets, etc.

Optimization is very good in TensorFlow. There are many opportunities to do hyper-parameter training. 

I don't have too much experience with the dashboards in the solution, however, it's possible they could be improved.

I need to have more experience in the security aspect of the solution. It could, however, always develop this area more.

It would be nice if the solution was in Hungarian. I would like more Hungarian NLP models. 

I've used the solution in the past 12 months. 

I haven't had any issues with stability so far. It's really reliable. There aren't issues with bugs or glitches or crashing.

The solution is absolutely scalable. My understanding of scalability is that when it comes to the solution, the learning task should run on the selected CPU. If I know how it should be and how it should be run, it's very easy as TensorFlow can also run on one CPU core or even on a GPU and so on.

Colab is great when I would like to learn something. You end up using Colab a lot. I like Jupyter Notebook and use it to create TensorFlow models.

There's also a lot of good documentation you can use to reference things and learn about the solution.

I have a bit of knowledge of PyTorch. I haven't used it much, however. I'm learning a bit about it now. While I don't have practical experience, I am taking a Coursera course that uses it.

In university, maybe ten years ago, I might have used something in MATLAB. I didn't have too much experience or too much knowledge about deep learning at that time, so I cannot say if it's hard or easy to do tasks in MATLAB or to compare the two. 

The initial setup isn't too complex. It's pretty straightforward. There is a lot of good documentation. There are many good courses. There are many good books from professors - from Ph.D.'s to data scientists. It's very, very easy. It's not so complex.

The deployment didn't take a very long time. In my experience, it only really took a few days. That is if a baseline model is enough for the client. Of course, if the requirement is an optimized model, it can be weeks or even a month.

The data processing, hyperparameter tuning, CPUs, and GPUs are all very, very important. If I have a very, very strong machine, I can do everything very fast, and it's a huge help for me.

I don't pay for the solution.

It's my understanding that if you want technical support you need to pay.

I'm just a user. I'm not a reseller or consultant.

I'm learning TensorFlow so I would like to do a TensorFlow certificate from Google in January or February. I'm learning how to deploy Poker with TensorFlow. It's new territory for me, however, it is very important.

I'm not sure which version of the solution I'm using. I have more developed servers and I'm using different versions.

I can recommend TensorFlow to anybody that wants to create deep learning models.

I'd rate the solution ten out of ten. I've been quite happy with it so far. 

I use this solution out of personal interest — for AI.

Our company doesn't use it too much. We tend to use OpenML, which is included in the OpenCV package, OpenML. 

It's got quite a big community, which is useful.

I tend to find it to be a bit too much orientated to AI itself for other use cases, which is fine — that's what it's designed for. Personally, I find it to be a bit too much AI-oriented.

We have been using TensorFlow for roughly five years.

We haven't had any issues stability-wise or scalability-wise. I tend to use it with Python. It seems okay. It works fine. 

The initial setup was straightforward.

I would happily recommend TensorFlow to people who are looking to use it for AI. Overall, on a scale from one to ten, I would give this solution a rating of eight.