December 07, 2021

It's a Golden-rumped Sengi! - Observation of the Week, 12/7/21

Our Observation of the Week is this Golden-rumped Sengi (Rhynchocyon chrysopygus), seen in Kenya by @mwas!

Mwangi Gitau grew up in Nakuru, Kenya, about ninety miles west of Nairobi, and says he first became interested in nature when he saw tourists “admire and take photos of the huge flocks of the Lesser Flamingos and other water birds” in Lake Elementeita - a saline lake in Africa’s Great Rift Valley. 

Now he leads birdwatching tours for ConQuest Adventures in Kenya, Uganda and Tanzania, which is where he photographed the observation of the week. 

The Golden Rumped Sengi was seen in Arabuko Sokoke Forest in Coastal Kenya, during an 8 days birding tour…We were on a trail looking at a Red-capped Robin-Chat, [and] deep in the thickets we could hear noises in the leaf litter. We stayed still for a while, and suddenly the sengi showed up on the trail - that’s how we managed to get the photo. The  Red-capped Robin-Chat follows the sengi when it's feeding because it disturbs insects on the litterfall, which the robin benefits from.  

Most of the clients who visit the forest always include it on their wish list, and luckily we saw close to 10 pairs in a transect of 12 miles for a duration of 3 days.

As its common name suggests, this species of sengi (also known as “elephant shrews”) has gold-colored fur on its hindquarters. This is thought to attract a predator’s attention to that part of its body, which has tougher skin and might deflect teeth or fangs - if they can even get close to this quick animal. Most of their diet consists of various invertebrates, which, as Mwangi explained, they find by digging in the leaf litter. The IUCN lists them as Endangered due to habitat loss and trapping.

Mwangi (above) tells me he’s been on iNaturalist for about three years after being persuaded to join by Mike Plagens (@mjplagens) and says “so far it’s a very good learning platform because you network with experts in all fields.”


- You can check out Mwangi’s bio here!

- Disovery/BBC shot some nice footage of a Rufous Sengi (Galegeeska rufescens).

Posted on December 07, 2021 22:31 by tiwane tiwane | 9 comments | Leave a comment

November 30, 2021

A Hazel Grouse Winter Portrait in Russia - Observation of the Week, 11/30/21

Our Observation of the Week is this Hazel Grouse (Tetrastes bonasia, Рябчик in Russian), seen in Russia by @zveroboy57!

Alexander tells me he’s been interested in nature since he was a child but has only been photographing wildlife seriously since 2013. If you check out the 500+ observations he’s posted to iNaturalist, you’ll see he’s captured many fantastic images of the animals around his home in central Russia.

And in that area, he says, Hazel Grouse are “not uncommon” and he’s often able to find them at watering holes and feeders. This species ranges throughout much of Eurasia and usually eats plants, although it also consumes insects when breeding. Measuring about 35–39 cm (14–15 in) in length, it’s one of the smaller members of the Family Phasianidae, which also contains chickens, peafowl, and pheasants.

Alexander (above, ready to photograph) joined iNaturalist just over three years ago and says “I upload to iNaturalist so that the world knows who lives in my area.”

(Google Translate was used to translate Alexander’s responses into English, and I lightly edited them for clarity.)

- Hazel grouse have a pretty high-pitched song, you can listen to it and watch some nice footage here.

Posted on November 30, 2021 20:45 by tiwane tiwane | 2 comments | Leave a comment

What are people posting to iNaturalist via Seek?

In the Spring of 2019, we added functionality for Seek by iNaturalist users to post to iNaturalist from Seek. To date, over 2.3M observations have been posted via Seek. For comparison, just under 100M observations have been posted to iNaturalist with 53.5M from the iNaturalist mobile apps and 41.9M posted via the website. Note these numbers include all observations not just Verifiable (where Verifiable = Needs ID & Research Grade). The graph below shows the accumulation of observations posted via Seek. Like other iNaturalist activity, there is a seasonal pattern that reflects the Northern Hemisphere summer.

Compared to the observations posted via the iNaturalist mobile apps, there is a noticeably larger fraction of captive observations have been posted via Seek. But the remaining quality grade categories are remarkably similar across both apps.

More plants and less birds have been posted via Seek compared with the iNaturalist apps. But again, otherwise the species categories are remarkably similar.

Proportionately fewer observations of rare (species with less than 50 observations), uncommon (species with 50-100 observations) and common (species with 100-1000 observations) species have been posted via Seek compared with the iNaturalist mobile apps. For reference, species in the computer vision model have around 100 observations or more.

Lastly, we compared the spatial distribution of observations posted via Seek and iNaturalist. The map below shows the ratio of observations posted via Seek to observations posted via the iNaturalist Mobile app in each 1-degree grid cell. Bluer colors indicate relatively more observations from Seek while red colors indicate relatively more observations from the iNat Mobile app. The Eastern US and Europe show up as Seek hotspots.

The observations posted to iNaturalist from Seek represent a small sample of Seek usage. But because the rest of Seek activity stays on users' devices and we know very little about it, it's the only insight we currently have. In summary, the kinds of observations are quite similar to what's posted through the iNaturalist apps with some interesting differences. There is a larger percentage of observations of non-wild organisms, more plants than birds. Likewise, observations of rarer species are posted less commonly through Seek and activity, relative to the iNaturalist mobile apps, is greatest in the Eastern US and Europe and a few other scattered enclaves like Peru.

Posted on November 30, 2021 18:24 by loarie loarie | 32 comments | Leave a comment

November 24, 2021

In Brazil a Beautiful Moth Becomes a House Visitor - Observation of the Week, 11/24/21

Our Observation of the Week is this stunning moth, likely in the genus Ernassa, seen in Brazil by @aninhafonte!

A visual artist, photographer and teacher who’s finishing a master’s degree program at a Federal University in Visual Art Education, Ana Carolina da Fonte tells me that has been in love with nature for most of her life. Growing up near the Atlantic Forest Reserve “[I] used to collect dead insects I found interesting,” she says, “mostly beetles, moths and butterflies. Back then there were no cell phones or digital cameras, [so] nowadays I am happy to register them with my phone or camera, but I still collect some here and there!”

That includes the remarkable moth you see photographed above, which actually appeared in her house at night, attracted by lights. 

My daughter loves moths, but she's also scared of them. This one happened to be inside her bedroom. Of course, I couldn't send it back to the forest before registering a few photos. It is beautiful, but I must say it's way more [beautiful] in person! The flash faded its colors a little and didn't do justice in the reds in its fuzzy body.

A member of Tribe Arctiini (commonly known as Tiger Moths), members of the genus Ernassa are found in neotropical regions of South America but I couldn’t find much other information about them. If you know anything, please add it as a comment! Many adult tiger moths are visually striking and they can be found almost anywhere around the globe. 

“I was happy when I found out about iNaturalist, because now I get to share all my findings with the world,” says Ana (above).

I also love learning about the new creatures I find and posting photos of them in the platform. Not only that, but I also check what other people have been posting near me, curious to what I might be finding next. Using iNaturalist changed the way I interact with nature, since it made me a little more obsessed in registering everything I see, because now I get to share with others and that makes it way much more fun!


- You can take a look at Ana’s Instagram feed here!

- National Moth Week has some helpful tips for finding moths!

Posted on November 24, 2021 23:59 by tiwane tiwane | 6 comments | Leave a comment

November 23, 2021

Meet Jazz (turtlehelper), an iNaturalist Monthly Supporter

This is the fourth interview in a series getting to know members of the iNaturalist community who are also Monthly Supporters. You can also read the first, second, and third interviews.

Jazz (@turtlehelper) from Kitchener, Ontario, Canada is passionate about conserving turtle species at risk and has been a part of many turtle data collection projects for the last 10 years. She can often be seen walking along busy roads adjacent to wetlands, moving turtles across roads, tracking mortalities, transporting injured turtles, and observing nesting sites. Jazz enjoys being outdoors doing fieldwork, but also taking opportunities to interact with the scientific community to educate herself and others. Currently, she is in the Herpetology Internship Program at Scales Nature Park (near Orillia).

Jazz with a Northern Map Turtle near Georgian Bay

How did you first get into iNaturalist?
My love of Ontario’s turtles led me to seek out volunteer opportunities over the years, but one particular project introduced me to iNaturalist. In 2019, several volunteers and I joined a local road ecology project which involved completing walking surveys and collecting data. Our job was to track all living and deceased wildlife found on a specific section of road, at certain times of the day. For this project we used iNaturalist on our smartphones, then added our geotagged pictures and other notes to the assigned project. Our data was used for road mortality and mitigation studies and to track populations of species-at-risk turtles and snakes. I had a positive first impression of using iNaturalist. Previously, volunteer projects required additional equipment and data was written out manually then entered into an online form. Using iNaturalist was a quicker process and more convenient!

Since contributing to that first project using iNaturalist, I have contributed data to many other projects, tracked new turtle populations on roads and in wetlands, learned about adjacent flora and fauna, and made regional projects (like this one for Waterloo) to help others get excited about species ID in their respective areas.

What made you want to donate monthly, in addition to everything else you do with iNaturalist?
I’ve had iNaturalist for a few years now, and enjoy using it on a regular basis because it keeps me curious and connected with nature. I donate monthly because I believe in the work being done and see iNaturalist as a useful tool for the world. For one, it helps people engage with their environment, learn species IDs, study the information, and interact with those who are more knowledgeable. iNaturalist is a forum to exchange learning; a bridge between scientists and the community. Second, iNaturalist is a worldwide collection of data that can potentially be used for positive changes in the natural world. iNaturalist needs the ongoing support, so all this work can grow and continue. Some days I collect lots of data, so I would like to at least cover my share of space on the server, if not donate more one day when I can afford it.

What keeps you motivated?
Having a target species to look for
I focus my efforts on turtles, building a repertoire of places to visit, and observing their behaviour in their habitat. I’m never done learning about them! I also learn about plants growing around the wetland where the turtles live, or notice other species that live nearby. It’s motivating to have something specific when you are out in nature.

Studying field guides and cross-referencing with other nature apps.
I own lots of physical guides for local herps, birds, plants etc. and other apps that I can cross reference for information when I go out. For example, I have apps that can play frog or bird calls and a guide that shows the differences between frogs that look similar. Of course, iNaturalist users can help me with IDs, but having my own references beforehand helps me to make informed guesses.

Joining meaningful projects and events on iNaturalist
I am motivated to contribute to projects that need data on species-at-risk needing protection. I seek out projects that help me learn more about the life and habitat of Ontario’s reptiles and amphibians. I have also been motivated by joining local Bioblitzes. These are opportunities I like because you are intentionally looking for certain species in a set area.

Just going out and looking for something new
When I am not looking for turtles, I look for any species I have never seen before, especially if visiting a different area. I look for odd shapes, bright colours, different textures, etc. When I feel unmotivated it gives me something to “hunt” for when using iNaturalist.

Interacting with others in person (consider pandemic rules first!)
I joined a nature club because it adds a social aspect you just can’t get the same way online. Also, going in the field with experts makes my time more enjoyable and learning is more nuanced. When someone shows me a species in person, it gives me more confidence for the next time I encounter it in the wild.

What’s something that you’d like more members of the iNaturalist community to know or do?
Learning should be a positive experience. When correcting others, try to be friendly and patient with people. Similarly, it’s better to ask questions rather than jump to conclusions. Give helpful answers to get others excited about what they are learning!

Thank you to @turtlehelper and all of the Monthly Supporters! We’re grateful to everyone who is generous with their time, expertise, and other gifts. iNaturalist Monthly Supporters give automatic, recurring charitable donations and can be recognized on their profile pages, if they choose to from their account settings. Monthly Supporters are a critical part of our community and help ensure that iNaturalist is freely available to people all over the world. You can become a Monthly Supporter by giving your first recurring donation online. Thank you!

Become a Monthly Supporter

iNaturalist is a joint initiative of the California Academy of Sciences and the National Geographic Society. All donations will be received by the California Academy of Sciences, a 501(c)(3) tax-exempt not-for-profit organization based in the United States of America (Tax ID: 94-1156258). Gifts can be made online in more than 40 different currencies via bank account, credit/debit card, or PayPal.

Posted on November 23, 2021 14:15 by carrieseltzer carrieseltzer | 4 comments | Leave a comment

November 18, 2021

Identifier Profile: @beartracker

This is the sixth in an ongoing monthly series profiling the amazing identifiers of iNaturalist! 

Oftentimes the only way most of us can regularly observe elusive wildlife like mammals is by finding traces of their passage: tracks, scat, rubs, and the like. Thankfully there are identifiers on iNat like @beartracker who can help with identification of those signs!

“I got my start in tracking when I was about 13 or 14,” says Kim Cabrera, who grew up in southern California. “My dad was the first one to point out tracks to me, when I was just a youngster. For some reason, those little imprints in the dirt just fascinate me. I imagine the animals making them and am curious about how those animals survive. As a teen, I started to teach myself the various tracks by drawing the ones in the field guide over and over.”

She continued to learn more about tracking and even learned to track humans for search-and-rescue purposes “[but] tracking animals is my first love.” After moving to the northern reaches of California over thirty years ago, she now tracks the bears and mountain lions in the area (among other animals), trying to determine their dietary and behavior habits from their tracks, signs, scrapes, and such. 

A CyberTracker-certified Track and Sign Specialist in both the forest and desert biomes, Kim also spends a good amount of time identifying track and scat observations on iNaturalist, which she discovered through a friend. 

I find it a great learning tool since you get to see tracks and sign in various environments and from all over the world. There is a very helpful community of trackers on iNaturalist who get into detailed discussions on identifications. We tag each other and discuss various tracks amongst ourselves and help each other learn. 

When adding IDs on iNat, Kim searches for observations of tracks and other sign and in addition to the identification she tells me

I try to remember to give reasons why I think a track is this not that so that it’s a learning experience for all. For the tracker, being able to articulate the reasons why you identified a track as you did helps you to clarify the details and solidifies it in your mind. So, the learning goes both ways. Hopefully, my track identifications are helpful to the folks who posted the images. Sometimes, it leads [to] a long discussion between folks, which can be helpful. 

For me, the thing about identifying things on iNaturalist is that it increases my knowledge. Tracks can appear in many different ways. For example, raccoon tracks have a reputation among trackers since they can mimic the appearance of those of other species. I have seen raccoon tracks that look just like those of bobcats or dogs!...

I think this has real-world applications. It’s not uncommon for mountain lions to be suspected of depredations on domestic animals. However, often the tracks found turn out to belong to dogs. Sadly, this can have really bad consequences for the mountain lions if the tracks are misidentified. So, the real-world application of tracking is my main motivation to help make correct track identifications, both on iNaturalist and other places. 

In addition to identifying, Kim also uses iNat as a place to post her observations of other organisms - she’s particularly interested in galls and leafminers at the moment and, after reading about them, she’ll comb through existing observations on iNat as well as post her own observations of them for ID help. “There are experts on feathers, bones, skulls, galls, and more, all on iNaturalist,” she says. “I have learned so much from them over the years and am grateful to them.”

I think learning is the most important thing we can do each day. Try to learn something new each day and you will never be bored. Learning about nature is my passion. I’ve discovered fascinating things with amazing life histories, like galls! And seeing these and identifying them on iNaturalist has helped open up a whole new world of knowledge to me. I am no expert in galls, but I do love learning about them. It’s the helpful identifications of those who ARE the experts that are helping me along that learning journey. As a specialist in track and sign, I hope that my knowledge of my particular subject area can prove useful to someone else who is as eager to learn as I am. This is why I love iNaturalist and its wonderful community. 


- Check out Kim’s website, as well as her YouTube channel!

- Kim also created the Animal Tracks and Signs by Beartracker Wildlife Tracking and Bird Tracks and Signs projects on iNat.

- References Kim recommends include books by Mark Elbroch as well as those by Moskowitz, Tkacyzk, Eiseman, Resendez, Poppele, McFarland, Lowery, and Halfpenny.

- Take a look at the printable comparision sheet Kim made for differentiating feline and canine tracks.

- if you have suggestions for identifiers to profile, please message @tiwane (don't write in the comments). I'm particularly interested in people who specialize in underrepresented taxa, regions, or types of observations.

Posted on November 18, 2021 22:22 by tiwane tiwane | 51 comments | Leave a comment

November 09, 2021

Meet the First Cladomelea debeeri Bolas Spider Posted to iNat! - Observation of the Day, 11/9/21

Our Observation of the Week is the first Cladomelea debeeri Bolas Spider posted to iNat! Seen in South Africa by @suncana.

Born in Croatia, Suncana Bradley says she’s always appreciated nature, “but moving from Croatia to South Africa about a decade ago opened up my eyes. The incredible biodiversity of the subtropical climate was the perfect motivation to start exploring and learning. I very quickly went from a casual appreciator to a keen naturalist.” She’s particularly into photographing insects (especially caterpillars) and is a member of the Caterpillar Rearing Group Africa, a group dedicated to learning the life histories of African Lepidoptera. 

And it was while out looking for caterpillars that Suncana spotted the spider you see photographed here. “Finding [Cladomelea debeeri] was pure luck,” she tells me. 

I was out in the bush in the early evening looking for caterpillars when I came across this odd looking spider. It was unlike anything I've ever seen, so I immediately posted the photographs on our local spider group. She was very quickly identified by a bunch of very excited spider experts, and a couple days later visited by John Roff who described the species in 2004. She is the first record of a sub-adult female, and we are hoping a male will come once she is mature. A male of the species has never been seen, so we are all quite excited. The species conservation status is currently vulnerable due to its limited range, so this is indeed a very special find.

Bolas spiders are so named for their hunting method. Rather than construct a web for trapping prey they use their legs to spin a piece of silk with a globule of sticky liquid on it (see above), similar to the bolas made famous by South American gauchos. Remarkably, they emit moth sex pheremones to lure male moths within striking distance. As Suncana said, this species was described in 2004 by John Roff and Ansie Dippenaar-Schoeman. Roff’s friend Len de Beer assisted in the collection of the type specimen, hence the spider’s scientific name. 

Suncana (above) uses iNat for her research and conservation efforts (she finds it especally valuable for getting her caterpillar’s host plants identified) and it provides a way for she and others to present the diversity and value of green spaces.

To me iNaturalist is more than that, though. It's an incredible fountain of collective knowledge that I have not come across anywhere else. I always naively thought that we know everything there is to know, what with it being the 21st century and all. This platform and its users have shown me that there is in fact more that we don't know, which just made me more passionate about learning and discovery. There is always something new and exciting to discover (my Bolas being the perfect example), so I keep striving for more. Exploring Africa wouldn't be half as fun without iNat.


- Not only did she take stills, but here’s Suncana’s footage of Cladomelea debeeri swinging her bolas.

- Suncana talks about her caterpillar rearing passion in this short video.

- Sir David Attenborough and team have some nice footage of a bolas spider catching prey.

Posted on November 09, 2021 21:41 by tiwane tiwane | 19 comments | Leave a comment

November 02, 2021

The Jumping Spider and the Blunthead Slug Snake - Observation of the Week, 11/2/21

Our Observation of the Week is this Blunthead Slug Snake (Aplopeltura boa) with a Jumping Spider (Family Salticidae) on its head! Seen in Indonesia by @fhadlikennedi.

Umar Fhadli Kennedi graduated with a degree in Conservation of Forest Resources and Ecotourism from IPB University in Bogor, Indonesia, back in 2018, and credits his days as a university student with his interest in nature. “When I was a kid,” he tells me, “I really liked outdoor activities such as hiking, but I wasn’t very aware of the plants and animals around me.”

But as a university student, I studied in the faculty of Forestry in Bogor, doing a lot of fieldwork, learning about a few species of plants and animals. I also joined a student club for herpetology, and I learned a lot about amphibians and reptiles there, where I also met the Indonesian herpetologist [@mirza] who made me want to learn more and more about amphibians and reptiles of Indonesia. I’m currently interested in amphibians and reptiles, but sometimes I also take a picture of insects and mushrooms when I'm in the field (but I don't even know what the species is, just upload in inaturalist haha).

This past June, Kennedi was collecting data for his supervisor’s research at Mount Halimun Salak National Park but no cellular signal reached their lodge. So, most nights he and his colleagues (below) hiked about fifteen minutes up to a hilltop in the park to connect with the world. “I always brought my camera with me, just in case we found something interesting,” he says. 

While up on the hill we looked for amphibians and reptiles and one night, one of my friends found this snake and started photographing it. As I was taking photos a spider jumped from the tail to the head of the snake. Seeing that, I shot as fast and as much as possible, hoping to get a great picture. The photos turned out too dark, haha, and I'm struggling to make the picture better, but I'm still not satisfied with the result.

As its common name suggests, the blunthead slug snake tends to prey on gastropods, but it will also eat lizards and other prey. Some species in the slug snake family (Pareidae) have quite asymmetrical jaws, with significantly more teeth on the right mandible than the left, allowing the snake to more easily pull most snails from their shells. However, not all snail shells curve to the right (dextral) - a minority curve to the left (sinistral), and it’s thought this has created pressure for snail speciation where some of these snakes occur. “Already,” says Andrew M. Durso, “southeast Asia harbors more sinistral snail biodiversity than any other region (12% as opposed to 5% worldwide), likely in part due to selection against dextral and for sinistral shells from snake predation.” These snakes are not venomous and, as far as know, harbor no specific attraction to jumping spiders. ;-)

Kennedi (above) posts his reptile and amphibian observations to his supervisor’s Amfibi Reptil Kita (ARK) project and tells me that at first he only used iNat for herp photos but 

after a few years spent in working with wildlife, I wanted to share other unique wildlife that I found (such as insects and mushrooms). Also, the iNaturalist community is so great that, even though I'm a newbie for most things beside herpetofauna, almost always some people help identify the species that I don't know.

(Quotes have been lightly edited for clarity)


- I linked to it in the text, but I wanted to make sure you check out Andrew M. Durso’s blog post about slug snakes and snail chirality.

- Here’s some nice footage of a blunthead slug snake from Penang, Malaysia.

- This video has a nice section showing a slug snake in Taiwan attemping to eat a sinistral snail, starting at around 29:34.

- This is not the first Observation of the Week documenting an arthropod in the head of a snake!

Posted on November 02, 2021 20:34 by tiwane tiwane | 15 comments | Leave a comment

October 26, 2021

A Bone-Chilling Stream in China Leads to a Torrent Frog Photo - Observation of the Week, 10/26/21

Our Observation of the Week is the first Green-spotted Torrent Frog (Amolops viridimaculatus, 绿点湍蛙 in Chinese (simplified)), seen in China by @jianwang!

A herpetologist at the Museum of Biology Sun Yat-sen University in Guangzhou, China, Jian Wang’s work focuses mostly on snakes, lizards, and anuras (frogs and toads) and he’s interested in both their taxonomy and conservation. Since he started his work in 2014, he’s “participated in the description of nearly 40 new species and the revision of several species groups,” and you can see his publications here. But this frog was seen far to the west, in China’s Yunnan Province. 

Yunnan Province has the highest biodiversity among all administrative regions in China,” he explains.

Located in the western Yunnan, the ample monsoon rainfall results in a rich diversity of fauna and flora in the Gaoligongshan Region (高黎贡山). Scientists have conducted biodiversity surveys since the 19th century, and plentiful new species/species with restricted distribution have been discovered. With over 700 recorded vertebrates, Gaoligongshan has become one of the places I’ve most yearned to visit…Herpetofauna of the northern, middle and southern parts of Gaoligongshan are distinctly different due to Gaoligongshan’s wide geographical expanse.

In 2015 he was finally able to make it there, which is where he came across the green-spotted torrent frog in its type locality. “[This species] is mainly distributed in the southwest slopes of Gaoligongshan,” says Jian Wang. “While I was investigating in a bone-chilling stream, this fascinating creature occurred, which escaped and allowed me to make only one shot.”

Like other members of its genus, this species does seem to prefer fast-moving cool streams in mountains and hills, and as tadpole it has a suction cup on its belly to help it stick to rocks. It’s also found in parts of India, Vietnam, Myanmar, and possibly Laos. The IUCN lists it as Near Threatened, with major threats being agriculture, logging, and dams or other water management.  

“We made other fascinating discoveries on the same trip, and several new species were gradually described,” says Jian Wang (above). These include the Western Yunnan Music Frog (Nidirana occidentalis, 滇西琴蛙), the Fei’s Horn Toad (Megophrys feii, 费氏角蟾), and the Banna Parachute Gecko (Gekko bannaense, 版纳伞虎)! 

“I love all other groups of animals in the world that that iNat gives me the opportunity to read about,” he says, “[and] I'm glad to help in any way I can in identifying amphibians and reptiles/sharing interesting observations.”


- @qin_huang, who also has also worked at Sun Yat-sen University, is the subject of a previous Observation of the Week post!

- This video shows a different Amolops frog (Amolops larutensis) in a fast-moving Malaysian stream.

Posted on October 26, 2021 21:59 by tiwane tiwane | 11 comments | Leave a comment

October 21, 2021

New Vision Model Training Started

We've started training on a new model, which will be our first model update since July 2021. Here's what you need to know.

It’s bigger

iNaturalist data continues to grow. This time around we went from 38,000 to 47,000 taxa, and from 21 million to 25 million training photos.





We’ve sped up training time again

iNaturalist has new computer vision hardware!. We have two more NVIDIA RTX 8000 GPUs, again granted to iNaturalist by NVIDIA. Based on early experiments, three GPUs seem to train about twice as fast as a single GPU in flat-out training speed. We also have a new computer vision server to house these GPUs, which has 4x the RAM, a hugely faster CPU, and really fast disks (at this scale, reading photos from disk and writing data back to disk is a limiting factor).

This training run is starting with the last checkpoint from the previous training run, rather than starting from the standard ImageNet weights like we did for the previous training run. Basically, this training run gets a head start in understanding what kinds of visual features are important for making iNaturalist suggestions.

We changed a few things about how we generate training data

Hybrid species are no longer included. The previous training run was the first time where we had a significant amount of training images for some avian hybrid taxa (mallard hybrids, for example), and including them as training categories really confused the model. Despite mallard being the most commonly observed species on iNaturalist, the most recent model had a hard time getting mallard suggestions right, struggling to tell the difference between mallards and the various mallard hybrids. Excluding hybrid species from the training set should keep the computer vision model on the task of trying to distinguish between visually distinct taxa.

The number of photos from an individual observation that could be included in the training set has been capped to a max of 5 photos per observation. Previously, in very rare instances, a single observation with hundreds of nearly identical photos could have dominated the training data for a single taxon, potentially causing the model to learn visual features from just those photos, to the detriment of generalizing well to others’ photos.

We changed one thing about how we train the model

Label smoothing is back in our training config. Label smoothing sets the “true” training labels to “softer” values like 0.9 instead of “harder” values like 1.0. Basically, when the model is shown a photo to learn from, we’re now saying “we’re very confident that it’s species X” instead of saying “we’re 100% convinced that it’s species X.” It’s designed to reduce overconfidence in model scores. This is something that we’ve done with some models in the past, but this configuration got lost in the transition for the previous training run.

When will it be ready?

This new model will take a few months to train, and then a few weeks to test / validate before we decide it’s ready for deployment. The increase in training images is slowing us down, but the new hardware is speeding us up. Starting from a checkpoint from our previously trained model also reduces the amount of training work that needs to be done.

Future Work?

The main priority is to get to our stated goal of training 2 models a year.

We're trying to be more transparent about when and how we train new models, so we’ll be working on changes for that. This post is a start, but you can expect a dashboard with more details and charts coming soon.

We have a grant from Amazon to explore ways to improve how we export and train our models, and we’ll be working on that before the end of the year.

Finally, we now have two computer vision systems, one dedicated to production training runs like this one, and another which we can use to run experiments to improve future training runs and explore other machine learning-based features for iNaturalist.

Posted on October 21, 2021 21:33 by alexshepard alexshepard | 31 comments | Leave a comment